Radar Data Analysis and Hidden Object Detection

Abstract

Techniques are described herein for analyzing radar data to determine whether radar noise from one or more target detections may conceal additional objects in the vicinity of the target detection. Determining whether an object may be concealed may be based at least in part on the radar noise level based on the target detection, as well as on a radar cross section and / or a distribution of Doppler data associated with a particular object type. For locations in the vicinity of the target detection, the radar system may determine an estimated noise level and compare the estimated noise level to a radar cross section probability associated with the object type to determine the likelihood that an object of the object type may be concealed at the location. Based on the analysis, the system may determine a vehicle trajectory or otherwise control the vehicle based on the likelihood that an object may be concealed at the location.

Description

[Technical field]

[0001] This disclosure relates to radar data analysis and hidden object detection. [Background technology]

[0002] CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to U.S. patent application Ser. No. 17 / 384,430, filed July 23, 2021, and entitled "RADAR DATA ANALYSIS AND HIDDEN OBJECT DETECTION," the entire contents of which are incorporated by reference into this specification for all purposes.

[0003]

[0002] Radars generally measure the distance from the radar device to the surface of an object by transmitting radio waves and receiving reflections of the radio waves from the surface of the object, which distance can be read by a sensor in the radar device. The sensor can generate a signal based at least in part on the radio waves incident on the sensor. This signal can include return signals due to reflections, but the signal can also include portions due to noise and / or other interfering signals (either from the radar device itself or from external sources). To distinguish the return signals from noise or other interfering signals, radar devices generally use a detection threshold to suppress false positive detection (i.e., identifying a portion of a signal as a return when in fact a portion of the signal is due to noise and / or other interfering signals). However, in some situations, such suppression can be problematic (and dangerous) since it can be associated with actual returns. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] U.S. Patent Application Serial No. 16,407,139 [Brief description of the drawings]

[0005] The detailed description will be set forth with reference to the accompanying drawings, in which the leftmost digit(s) of a reference number identifies the drawing in which that reference number first appears. Use of the same reference number in different drawings indicates similar or identical items or features.

[0006] [Figure 1]

[0004] FIG. 1 illustrates an example scenario in which a radar device of an autonomous vehicle receives and analyzes radar data including data indicating that an object may be obscured by radar noise associated with the detection, in accordance with one or more embodiments of the present disclosure. [Diagram 2] FIG. 2 is a diagram illustrating an example radar cross section distribution and an example Doppler distribution associated with a pedestrian object type, in accordance with one or more embodiments of the present disclosure. [Diagram 3] FIG. 3 illustrates an example radar data including pedestrian-based detection in an environment, estimated noise levels, and radar return signals in accordance with one or more embodiments of the present disclosure. [Figure 4]

[0007] FIG. 4 illustrates an example grid map of an environment associated with an autonomous vehicle, including indications of drivable and non-drivable areas based on radar noise levels and object type specific thresholds, in accordance with one or more embodiments of the present disclosure. [Diagram 5]

[0008] FIG. 5 is a graph showing the distance and Doppler delta between a detection and an object close to the detection in accordance with one or more embodiments of the present disclosure. [Figure 6]

[0009] FIG. 6 is a flow diagram illustrating an example process for determining drivable and non-drivable surfaces and controlling a vehicle based on radar data analysis in accordance with one or more embodiments of the present disclosure. [Figure 7]

[0010] FIG. 7 illustrates a radar data model for determining the probability of an object being hidden at a position based on Doppler range probability and a corresponding radar response threshold in accordance with one or more embodiments of the present disclosure. [Figure 8]

[0011] FIG. 8 is a flow diagram illustrating another example process for determining drivable and non-drivable surfaces for controlling a vehicle based on radar data analysis in accordance with one or more embodiments of the present disclosure. [Figure 9]

[0012] FIG. 9 is a block diagram of an example system for implementing the various techniques described herein. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0007]

[0013] The techniques discussed herein relate to analyzing radar data generated by a vehicle moving through an environment and determining that radar noise based on one or more target detections may potentially conceal additional objects in the vicinity of the detected target. As described in various examples below, the radar system may determine a radar noise level based on a side lobe level associated with a target detected in the environment. Radar response thresholds, including radar cross section (RCS) thresholds and / or Doppler thresholds associated with a particular object type, may be determined based on a radar response distribution associated with the object type. For example, one or more radar response thresholds may be calculated based on a distribution from a pedestrian object type and a predetermined probability and / or confidence threshold (e.g., 95%) associated with detecting a pedestrian in the radar noise. For a particular location (or region) near a target detection, the radar system may determine an estimated noise level for that location and compare the estimated noise level to the RCS and / or Doppler thresholds associated with one or more object types to determine the likelihood that an object of the object type may be concealed at that location. Based on analysis of the radar data, the vehicle may determine a trajectory or otherwise control the vehicle based on the likelihood that one or more objects may be hidden at the location.

[0008]

[0014] Analysis of radar data to detect objects may include techniques for reducing or suppressing false positive detections. A false positive detection may include a determination of an object detection in the radar data that is caused by radar noise rather than a physical object in the environment. Conventional techniques for radar false positive detection suppression include determining a detection threshold to reduce the number of false positive detections and / or keeping the number of false positive detections below a specified false positive detection rate. For example, a constant false alarm probability technique (CFAR), such as cell averaging CFAR (CA-CFAR), may test each cell (or portion) of radar data to determine whether a signal associated with the cell of interest meets or exceeds a detection threshold determined by averaging the signal of the cell of interest and / or signals associated with cells around the cell of interest. Such techniques assume that if the cell of interest includes a return signal (e.g., radar response data indicative of a detected object), the surrounding or nearby cells are likely to provide a good estimate of the radar noise in the scanned area. Other techniques may be used alternatively or in addition to CFAR techniques, such as selecting the largest portion of the signal associated with the cell as the return signal.

[0009]

[0015] However, application of false positive detection suppression techniques to radar data may tend to increase the number of false positives, where response signals caused by physical objects detected in the environment are due to radar noise and are suppressed. For example, an estimate of radar noise in an area of ​​the environment may mask radar detection of smaller and / or less radar reflective objects below the noise threshold. In the case of an autonomous vehicle, false radar detections, which represent a failure to identify physical objects in the environment, may pose significant risks to the safety of the vehicle, its occupants, and the surrounding environment.

[0010]

[0016] To address the technical problem of radar false positive and false negative detection, the techniques discussed herein may include determining thresholds based on radar response distributions of specific object types to determine drivable and non-drivable surfaces for vehicles in an environment. As described in various examples below, a radar noise estimator may use the locations and attributes of targets detected in the radar data to determine estimated radar noise levels at other locations near the detected targets. An object type threshold component may determine various thresholds (e.g., RCS and / or Doppler thresholds) associated with specific object types (e.g., pedestrians) that may be compared to noise levels at various locations in the environment to determine drivable and non-drivable surfaces.

[0011]

[0017] A non-drivable surface may be an area / location in the environment where the radar noise level is high enough (e.g., meets or exceeds a threshold) such that the area potentially conceals one or more objects. Although a physical object may or may not be present at the non-drivable surface location, the radar noise may effectively render the non-drivable surface unscannable by a radar device. Thus, the vehicle's control and navigation system may assume that each location of the non-drivable surface (e.g., one or more radar cells) contains an object, even if the presence of the object cannot be verified. In contrast, a drivable surface may be an area / location in the environment where the radar noise level is below a threshold, indicating that the radar noise is capable of concealing an object with a predetermined probability or confidence. As described in more detail below, the threshold used to determine drivable and non-drivable surfaces may be associated with a particular object type (e.g., pedestrians) and a predetermined probability based on the object distribution. For example, the threshold may be a value determined to include 95% radar response from pedestrians in the environment. The different thresholds may be different values ​​determined to include a different percentage of pedestrians in the environment (e.g., 90%, 99%, etc.) or a predetermined percentage of radar response from different object types (e.g., bicycles, cars, animals, etc.) As described in more detail below, the object type specific radar response thresholds may include a signal power value (e.g., an RCS value or other indicator of radar return power or strength), a Doppler value, or a combination of power and Doppler values.

[0012]

[0018] In some examples, the radar response threshold associated with an object type may be determined based on a predetermined probability value for objects of the object type. For example, based on a desired probability of 95% of a pedestrian, a pedestrian RCS distribution may be used to determine a corresponding RCS threshold and / or a pedestrian Doppler distribution may be used to determine a corresponding Doppler threshold. In other examples, a predetermined threshold may be used to determine a corresponding probability (or confidence level) associated with a threshold. For example, based on an RCS radar response threshold of -50 dB, a pedestrian RCS distribution may be used to determine a percentage of pedestrians that exceed the threshold. The percentage of pedestrians that exceed the threshold may also correspond to the likelihood that if a pedestrian is present at that location in the radar data, the pedestrian's RCS will exceed the threshold and that the pedestrian will be visible to a radar system that applies the threshold as a noise level threshold.

[0013]

[0019] The object type radar response thresholds, including the RCS threshold and / or the Doppler threshold, may also be modified upward or downward during operation of the vehicle to improve safety and / or driving efficiency of the vehicle as needed. For example, a component of the vehicle may adjust the radar response thresholds used for one or more object types to change the drivable and non-drivable surfaces determined for the vehicle based on the thresholds. For example, if the environment around the vehicle does not include sufficient drivable surfaces to allow the vehicle to move through the environment, the object type threshold component may increase the radar response threshold to increase the amount of drivable surfaces available to the vehicle. In contrast, if the environment includes more sufficient drivable surfaces, the object type threshold component may increase one or more radar response thresholds to increase the confidence level that a potential object could not be obscured by radar noise in the environment.

[0014]

[0020] As described in more detail below, in some examples, different thresholds may be defined for different Doppler ranges. In some examples, the estimated radar noise level of an area in an environment may be based on the difference in radar range between the area and a nearby target detection location, and may also be based on the difference in radar measurements (e.g., Doppler and / or RCS) between the area and the target detection location. For example, as the Doppler difference between the target detection location and another location increases, the estimated noise level at the other location may decrease. As a result, different radar signal power (e.g., RCS) thresholds may be determined and applied to different Doppler ranges. Different thresholds may be evaluated for different Doppler ranges and used with the associated object-specific probabilities of the different Doppler ranges to determine a total probability that an object of the object type may be hidden at the location (e.g., by summing the separate probabilities for the different Doppler ranges).

[0015]

[0021] In at least some examples, multiple determinations may be made for object types that vary in the radar data. In the case of radar data mapping, such as a grid of cells, a non-grid mapping, and / or a radar data contour, drivable and non-drivable surface maps may be determined for the vehicle for one or more different object types. For example, a drivable / non-drivable surface map based on potentially hidden pedestrians may differ from a map based on potentially hidden bicyclists, etc. In some examples, an autonomous vehicle may generate a trajectory for controlling operation of the autonomous vehicle based at least in part on one or more drivable / non-drivable surface maps in accordance with the techniques discussed herein. Additionally or alternatively, the autonomous vehicle may activate a collision avoidance system (CAS), a remotely operated computing device, and / or engage or disengage certain autonomous driving features based on the determination of the drivable / non-drivable surface map and / or potentially hidden objects, based on the techniques discussed herein.

[0016]

[0022] Further, in some examples, the techniques described herein may additionally or alternatively include determining estimated radar noise levels, object-specific radar response thresholds, and / or drivable / non-drivable surfaces based on object detection and other sensor data received from additional or alternative sensor modalities (e.g., cameras, LIDAR sensors, etc.). In some examples, the various techniques described herein (e.g., radar noise estimation of locations within the environment, object type radar response threshold determination, and threshold-based drivable / non-drivable surface determination, etc.) may be performed in response to a determination that other vehicle sensors, such as cameras or LIDAR sensors, may be occluded or blocked by steam, optical flares, reflections, etc. Additionally or alternatively, the techniques described herein may be performed in a first operation, after which any locations determined to potentially occlude pedestrians (or other object types) may be provided to additional operations using other sensor modalities to further analyze the locations (e.g., visual object recognition and analysis, etc.).

[0017]

[0023] Techniques described herein for determining locations within radar data where objects may potentially be hidden may additionally or alternatively include receiving and / or determining radar response profiles (or response profiles) associated with various object types. The response profiles may parameterize characteristics of the object types that affect how the object types affect radio waves and thus how the object types "appear" in the radar sensor output signal. For example, the response profiles may include a receive power and / or RCS associated with the object types. In some examples, the receive power and / or RCS values ​​indicated by the response profiles may be deterministic, or in additional or alternative examples, the values ​​may be probabilistic (e.g., indicated by a probability distribution function associated with the object type). In some examples, the likelihood that an object will be detected or will not be detected may be determined for one or more object types. For example, the techniques may include determining a first likelihood that a pedestrian will not be detected as a particular location, a second likelihood that a large vehicle will not be detected at the location, a third likelihood that a small vehicle will not be detected, a fourth likelihood that a traffic sign will not be detected, etc. In some examples, the techniques may include storing response profiles associated with object type, object size, object reflectivity, etc. In some examples, the response profiles may be indexed by range, azimuth, and / or Doppler.

[0018]

[0024] The techniques may additionally or alternatively include determining likelihoods associated with object types and portions of the environment (e.g., for different bins of range (or distance), azimuth, Doppler, and / or elevation). The techniques may include associating the likelihoods with portions of a radar spatial grid. For example, the radar spatial grid may include multiple cells, each of which may represent a different portion of the environment and / or a different bin of radar data. In some examples, a cell may have one or more likelihoods associated with it, and each likelihood may be associated with a different object type.

[0019]

[0025] In some examples, an autonomous vehicle may generate a trajectory for controlling the movement of the autonomous vehicle based at least in part on the radar spatial grid. The techniques may thereby improve the safety and effectiveness of the operation of the autonomous vehicle. Additionally, the techniques discussed herein may enable a computer to infer the presence of potential false negatives in radar data without having to receive raw radar signals and / or without (proprietary) information regarding the algorithm by which the radar device generates the detections.

[0020]

[0026] FIG. 1 illustrates an example scenario 100 including an autonomous vehicle 102 configured to determine drivable and non-drivable surfaces in an environment using radar response thresholds associated with objects. In some examples, the autonomous vehicle 102 may be an autonomous vehicle configured to operate according to a Level 5 classification issued by the National Highway Traffic Safety Administration, which describes a vehicle capable of performing all safety-critical functions for the entire journey in situations where a driver (or passenger) is not expected to control the vehicle at any time. However, in other examples, the autonomous vehicle 102 may be a fully or partially autonomous vehicle having other levels or classifications. It is contemplated that the techniques herein may be applied to more than just robotic control, such as autonomous vehicles. For example, the techniques discussed herein may be applied to airspace object detection, manufacturing, augmented reality, and the like. Additionally, even though the autonomous vehicle 102 is depicted as a land vehicle, in some examples, the autonomous vehicle 102 may be a spacecraft, a surface vessel, and / or the like.

[0021]

[0027] In accordance with the techniques discussed herein, the autonomous vehicle 102 may receive sensor data from sensors 104 of the autonomous vehicle 102. For example, the sensors 104 may include position sensors (e.g., global positioning system (GPS) sensors), inertial sensors (e.g., acceleration sensors, gyroscope sensors, etc.), magnetic field sensors (e.g., compasses), position / speed / acceleration sensors (e.g., speedometers, drive system sensors), depth position sensors (e.g., lidar sensors, radar sensors, sonar sensors, time-of-flight (ToF) cameras, depth cameras, and / or other depth-sensing sensors), image sensors (e.g., cameras), audio sensors (e.g., microphones), and / or environmental sensors (e.g., barometers, hygrometers, etc.).

[0022]

[0028] The sensors 104 may generate sensor data that may be received by a computing device 106 associated with the vehicle 102. However, in other examples, some or all of the sensors 104 and / or computing devices 106 may be separate from and / or located remotely from the autonomous vehicle 102, and data capture, processing, command and / or control may be communicated to the autonomous vehicle 102 by one or more remote computing devices over wired and / or wireless networks.

[0023]

[0029] FIG. 1 illustrates an example of a radar device 108 that may be associated with an autonomous vehicle 102 and / or may collect radar sensor data as the autonomous vehicle 102 moves through an environment. In some examples, the radar device 108 may have a field of view α that covers at least a portion of the autonomous vehicle 102's surrounding environment, i.e., a scan area. In various examples, the autonomous vehicle 102 may have any number of one or more radar devices and / or sensors (1, 2, 4, 5, 8, 10, etc.). In the illustrated example, the field of view α is expressed as approximately a 180-degree field of view. However, in other examples, the field of view of the radar sensor may be greater (e.g., 220 degrees, 270 degrees, 360 degrees) or less (e.g., 120 degrees, 90 degrees, 60 degrees, 45 degrees) than this angle. Additionally, the autonomous vehicle 102 may include multiple radar sensors with multiple different fields of view, ranges, scan rates, etc. In some examples, the autonomous vehicle 102 may include multiple radar sensors having at least partially overlapping fields of view such that the multiple radar sensors commonly capture at least a portion of the environment.

[0024]

[0030] The computing device 106 may include a perception engine 110 configured to determine what is present in the surrounding environment of the autonomous vehicle 102. Although not shown in FIG. 1, the computing device 106 may also include a prediction component including functionality for generating predictive information associated with the environment, and a planning component including functionality for determining how to operate the autonomous vehicle 102 within the environment based on information received from the perception engine 110. The perception engine 110 may include one or more machine learning (ML) models and / or other computer-executable instructions for detecting, identifying, segmenting, classifying, and / or tracking objects from sensor data collected from the environment of the autonomous vehicle 102. In some examples, the computing device 106 may also include a radar response threshold component 112, a radar noise estimator 114, and / or a threshold evaluation component 116. As described in more detail below, the radar response threshold component 112 may include functionality for determining a radar response threshold that may be applied by the autonomous vehicle 102 while traveling in the environment. The radar response threshold may be determined based on vehicle safety metrics, radar response distribution of object types, and / or drivable surface metrics, balancing the likelihood that an object may be obscured by radar noise in the area of ​​the radar data with the desirability of determining a safe and efficient route for the autonomous vehicle 102 to travel through the environment. The radar noise estimator 114 may be configured to determine an estimate of radar noise (e.g., RCS noise and / or Doppler noise) based on detection of an object by the radar device 108 in the autonomous vehicle 102. The radar noise received by the radar device 108 may be based on side lobe levels generated by an object reflecting a radar transmission signal. As described in more detail below, the radar noise at a location near the target detection may be based on a range difference between the target detection and the location, a Doppler difference between the target detection and the location, and a power (e.g., strength) of the target detection.The threshold evaluation component 116 may evaluate radar response thresholds (e.g., RCS and / or Doppler thresholds) associated with particular types of objects (e.g., pedestrians, bicycles, animals, cars, etc.) against estimated radar noise levels in different regions of the radar data to determine drivable and non-drivable surfaces for the autonomous vehicle 102.

[0025]

[0031] In some examples, the computing device 106 may receive radar data from one or more radar devices 108. For example, the radar device may receive a return signal (e.g., radio waves reflected from an object) based on a transmitted signal and may determine an estimated noise level based at least in part on characteristics of the return signal, such as average strength and / or average power. The characteristics of the return signal may depend on the hardware and / or software configuration of the radar device (e.g., may depend on the type of CFAR algorithm used by the radar device). In some cases, the radar noise level may be expressed as a constant noise floor, although additional or alternative noise floor types are contemplated as described further herein. In some examples, the radar device may output an object detection associated with a portion of the return signal that meets or exceeds the radar noise level. When outputting return radar signals and / or object detections, the radar device may output position data indicating the distance (e.g., range), azimuth (e.g., the scan angle at which the object was detected), Doppler, elevation, received power (e.g., the strength and / or power of the return signal), SNR, and / or RCS associated with the detected object. In various examples, the perception engine 110 may receive radar data that may include object detections, and may or may not include data regarding the original signal from which the object detections were derived and / or the raw signal itself.

[0026]

[0032] It should be noted that although the graphs and grids illustrated herein are illustrated as two-dimensional graphs and grids, the data represented thereby may be more than two-dimensional. For example, radar data and / or object detection may include one or more dimensions, such as, for example, received power, range, azimuth (e.g., the scan angle at which an object is detected), Doppler, elevation, received power (e.g., return signal strength and / or power), SNR, and / or RCS.

[0027]

[0033] In some examples, the computing device 106 may additionally or alternatively receive an image 118 from the sensor 104. The image 118 illustrates an example scenario in which a radar device may detect one or more targets and may receive radar noise that potentially masks additional objects from detection of the return signal. In general, the strength of the power of a radar return signal may depend on the size, material, orientation, and / or surface angle of the object that reflects the radio waves to produce the return signal. Thus, large objects such as the vehicle 120 and truck 124 shown in the image 118 may tend to generate larger return signals in the radar signal than smaller objects such as the pedestrian 122. In some examples, large return signals such as those produced by the vehicle 120 and / or truck 124 may skew the estimated noise level determined to be higher by the radar device due to the greater strength of the return signals due to the larger objects. Smaller objects (e.g., objects with return signals of lesser strength) in the same or similar area, such as pedestrian 122, may go undetected by the radar device due to the skewing noise threshold. The proximity to which an undetected object must approach a larger "skewing" object to be detected may depend on the technique at which the noise level is set by the radar device. For example, if the estimated noise level is based on the area of ​​a cell, smaller objects in that area may not be detected, but if the estimated noise level is determined on a cell-by-cell basis, the likelihood of detecting the smaller object may be slightly improved.

[0028]

[0034] In some examples, the computing device 106 may also include a radar response threshold component 112 configured to determine radar response thresholds for the autonomous vehicle 102 while traveling within the environment. In this example, the radar response threshold component 112 is implemented on the autonomous vehicle 102, but in other cases, the radar response threshold component 112 may be implemented on a separate computing device (e.g., computing device 938, described below). In such cases, the remote radar response threshold component 112 may determine and transmit radar response thresholds to one or more autonomous vehicles 102 for use during traveling operations using various techniques described herein. In yet other examples, some portions of the radar response threshold component 112 may be implemented external to the autonomous vehicle 102 to determine generalized data-driven radar response thresholds, while other portions of the radar response threshold component 112 may be implemented on the autonomous vehicle 102 to customize or tune the thresholds to a particular environment, driving conditions, user preferences, or the like.

[0029]

[0035] The radar response threshold component 112 may use one or more data items from a variety of different data sources to determine a radar response threshold to be applied by the autonomous vehicle 102. For example, as shown in this example, the radar response threshold component 112 may receive and use vehicle safety metrics 126 to determine the radar response threshold. The vehicle safety metrics 126 may include one or a combination of minimum acceptable driving safety criteria, such as accident rates (e.g., crashes per mile driven), injury rates, vehicle or property damage rates, estimated fatality rates, etc. Other types of vehicle safety metrics 126 may include near-miss collision rates, speed limit or traffic violation rates, excessive braking or acceleration rates, comfort metric compliance rates, and / or any data associated with the performance of the autonomous vehicle 102. In some cases, the vehicle safety metrics 126 may also be adjusted based on the amount of damage or injury likely to occur if a collision were to occur. For example, the vehicle safety metric 126 may be modified upward or downward based on the current speed of the autonomous vehicle 102 and / or the type of object the vehicle could potentially collide with (e.g., a pedestrian, bicycle or car versus a traffic sign or mailbox).

[0030]

[0036] When using vehicle safety metrics 126 to determine the radar response threshold, in some cases, the radar response threshold component 112 may evaluate the false negative and false positive detections (as well as true negatives and true positives) of multiple different possible thresholds, and the radar response threshold component 112 may select a threshold that minimizes the sum of the false negatives and false positives, or otherwise provide an efficient balance between the likelihood of failing to detect an object obscured by radar noise in a region of radar data (e.g., a false negative) and incorrectly determining that an object may be obscured when the object is not present in that region of the environment (e.g., a false positive).

[0031]

[0037] In some examples, the radar response threshold component 112 may evaluate a particular radar response threshold by applying the threshold to various ground truth driving scenarios. The ground truth driving scenarios may be simulated scenarios, log-based scenarios, and / or live driving scenarios. To evaluate the radar response threshold, the radar response threshold component 112 may apply the threshold during one or more ground truth driving scenarios. The radar response threshold component 112 may apply the threshold to regions of the radar data affected by radar noise from nearby object detections to determine whether the radar noise in the region is large enough to potentially hide another object (e.g., a positive object detection) or whether the radar noise in the region is insufficient to hide another object (e.g., a negative object detection). The radar response threshold component 112 may then use ground truth data, which may include data from other sensor means (e.g., lidar or image data) or manually labeled or verified data, to verify the positive or negative object detection and confirm whether an object is present in the region. Based on a positive or negative object determination using the radar response threshold and corresponding validation with ground truth data, the radar response threshold component 112 may identify that the determination is one of a true positive, true negative, false positive, or false negative. The radar response threshold component 112 may perform multiple such analyses of the radar response threshold in multiple different driving scenarios and combine the results to extrapolate to a number (or rate) of accidents, collisions, damage / injuries, or other vehicle safety incidents. Thus, the radar response threshold component 112 may determine a vehicle safety metric associated with a particular radar response threshold, or conversely, may determine a radar response threshold 132 to be applied by the autonomous vehicle 120 based on the provided target or desired vehicle safety metric 126.

[0032]

[0038] As mentioned above, the radar response threshold component 112 may also use the radar response distribution 128 in some examples to determine a radar response threshold 132 for the autonomous vehicle 102. The radar response distribution 128 may include a power distribution (e.g., an RCS distribution) and / or a Doppler distribution and may be associated with a particular object type (e.g., a pedestrian, a bicycle, a car, a truck, an animal, etc.). Using one or more radar response distributions 128, the radar response threshold component 112 may determine a radar response threshold based on a desired probability of an object of the object type that the threshold detects (e.g., 95% of a pedestrian).

[0033]

[0039] Additionally or alternatively, the radar response threshold component 112 may determine the radar response threshold 132 using a drivable surface metric 130. The drivable surface metric 130 may include a minimum amount and / or percentage of the area around the autonomous vehicle 102 that is designated as a drivable surface. For example, based on a predetermined drivable surface metric 130 of 60%, the radar response threshold component 112 may determine a corresponding radar response threshold 132 that results in 60% of the area (e.g., radar cells) around the autonomous vehicle 102 being drivable.

[0034]

[0040] In some examples, the radar response threshold component 112 may determine a radar response threshold 132 for the autonomous vehicle 102 based on a combination and / or balance between the vehicle safety metric 126 and the drivable surface metric 130 to provide both a high level of vehicle safety and driving efficiency through an environment. As described below, if a relatively large radar response threshold 132 is selected, the threshold evaluation component 116 may identify fewer regions of the radar data as potentially hiding objects in the radar noise, and thus have a greater overall drivable surface area. In contrast, if a relatively small radar response threshold 132 is selected, the threshold evaluation component 116 may identify more regions of the radar data as potentially hiding objects, and as a result, the overall drivable surface area is less. In some cases, the radar response threshold component 112 may determine a maximum radar response threshold 132 that meets a given vehicle safety metric 126 and a minimum radar response threshold 132 that meets a given drivable surface metric 130. In such a case, the radar response threshold component 112 may select a radar response threshold 132 between the determined minimum and maximum values ​​to meet both the desired vehicle safety metric 126 and the drivable surface metric 130.

[0035]

[0041] In some examples, the general radar response threshold may be determined by a remote computing system using techniques described herein and transmitted to the autonomous vehicle 102 where it may be modified based on the current driving environment. For example, the radar response threshold component 112 may tune the general radar response threshold upward or downward to determine the specific radar response threshold 132 based on the current route, traffic conditions, time of day, the number of pedestrians (or other objects) in the environment, etc. Based on the environmental conditions, the radar response threshold component 112 may determine that the radar response threshold should be decreased to increase the safety of the vehicle in the current environment, or that the radar response threshold should be increased to increase the amount of drivable surface in the environment.

[0036]

[0042] Additionally or alternatively, the general radar response threshold may be adjusted upward or downward based on the current driving environment and / or driving conditions of the autonomous vehicle 102. For example, the radar response threshold component 112 may perform a first modification to the general radar response threshold when the autonomous vehicle 102 is driving in a city, and a second modification to the general radar response threshold when the autonomous vehicle 102 is driving in a less dense rural environment. The radar response threshold component 112 may also modify the general radar response threshold based on detecting that the autonomous vehicle 102 is driving in inclement weather (e.g., rain, snow, fog, etc.). In some examples, the radar response threshold component 112 may also modify any vehicle safety metrics 126 based on the current driving environment and / or driving conditions of the autonomous vehicle 102.

[0037]

[0043] The radar response threshold component 112 may also adjust the radar response threshold 132 upward or downward for a particular region, or apply different object-specific thresholds to a region based on the type of road surface for the region. For example, if a radar data region potentially affected by radar noise corresponds to a sidewalk, the radar response threshold component 112 may apply a radar response threshold 132 configured to detect pedestrians. In contrast, for different radar data regions corresponding to different surface types (e.g., bike lanes, driving or carpool lanes, etc.), the radar response threshold component 112 may apply different radar response thresholds 132 configured to detect different object types (e.g., bicycles, small cars, etc.).

[0038]

[0044] In some examples, the perception engine 110 may determine the probability (or confidence level) that a radar return signal associated with a particular location is a true positive, a false positive, a true negative, or a false negative. For example, to determine various probabilities of a return signal at a location, the perception engine 110 may use the radar noise estimator 114 to determine an estimated noise floor, and the threshold evaluation component 116 to determine a radar response threshold associated with a particular object type (e.g., a pedestrian), and determine the probability of a hidden object at the location based at least in part on the estimated noise floor and the object type threshold. The radar noise estimator 114 may determine one or more estimated noise levels at the location based at least in part on one or more object detections received from the radar device, and the radar response threshold component 112 may use a distribution associated with a particular object type to determine RCS and / or Doppler thresholds associated with the object type and / or different object parameters (e.g., size, estimated reflectivity).

[0039]

[0045] Based on the estimated noise level at a location and the radar response thresholds for various object types, the perception engine 110 may determine whether the location is sufficiently likely to contain an object of the object type (e.g., meets or exceeds a probability threshold). In other words, the perception engine 110 may determine whether the radar device is capable of detecting an object of the object type in a particular region of the environment associated with the portion of the radar data based on the estimated radar noise and the object type threshold. In some examples, the portion of the radar data may be defined as a bin or some other portion thereof that represents a portion of the environment. A bin may include radar data associated with a set of distances, azimuth angles, elevations, and / or received powers. For example, a particular bin may include received power data associated with all object detections associated with the range of azimuth angles, distances, and / or elevations specified by the bin. In some examples, a bin may be associated with and / or represented as a cell in the radar space grid 134, as described in more detail below. Although a radar spatial grid 134 is used in this example, in other cases, radar data for the environment may be represented in other non-grid configurations, such as radar cells, radar-based contours, etc. In some examples, bins may be associated with set constants and / or ranges of values ​​in one or more dimensions of the radar data. For example, bins may be associated with a certain elevation angle, a range of distances, a range of azimuth angles, and / or any Doppler value.

[0040]

[0046] The perception engine 110 may additionally or alternatively utilize sensor data captured by the sensor 104 to generate and maintain one or more spatial grids, including a radar spatial grid 134, which may include cells associated with regions of the environment. Each cell of the radar spatial grid 134 may be associated with a bin (or other portion) of radar data. In some examples, the perception engine 110 may determine whether the radar data associated with the cell is likely to include a hidden object and may indicate whether the likelihood meets or exceeds a threshold associated with the object type. In some examples, a cell of the radar spatial grid 134 may alternatively or additionally include a designation as being occupied by an object (e.g., that at least one object detection has been output by a radar device or other system of the autonomous vehicle 102 in association with the region of the environment corresponding to the cell) and / or additional metadata associated with such detection, e.g., other data determined by the perception engine 110, such as semantic labels, lidar points, lidar instance segmentation, radar instance segmentation, etc., as described in more detail herein.

[0041]

[0047] In this example, the radar spatial grid 134 is shown as an 8×8 grid for simplicity. However, in other examples, any size (e.g., real-world area associated with the spatial grid), shape (e.g., length and width of the spatial grid), and resolution (e.g., size of cells used to make up the spatial grid) may be used for the radar spatial grid 134 depending on the accuracy and precision required, bin dimensions, memory size constraints, processing speed and load constraints, sensor range limitations, etc. In some examples, the spatial grid may be sized and shaped to match the reliable range of the sensor 104 used to capture the sensor data into which the spatial grid is input, and the resolution may be selected by maximizing the accuracy and precision required for a given application given memory and processing constraints. In some examples, the length and width may be the same, while in other examples, the length and width may be different. In a specific example, the size of the spatial grid may be approximately 50-200 meters wide and approximately 50-200 meters long, with a resolution of 0.25 meters per cell.

[0042]

[0048] In some examples, the perception engine 110 may additionally or alternatively determine the location of the autonomous vehicle 102 as determined by a location engine (not shown) that may use any sensor data to locate the autonomous vehicle 102, data related to objects in the vicinity of the autonomous vehicle 102 (e.g., classifications, instance segmentations, trajectories associated with detected objects), route data specifying the vehicle's destination, global map data identifying road features (e.g., features detectable in different sensor modalities useful for locating the autonomous vehicle), local map data identifying features detected in proximity to the vehicle (e.g., locations, positions and / or sizes of buildings, trees, fences, fire hydrants, stop signs, and other features detectable in various sensor modalities), etc. The data generated by the perception engine 110 (including the radar space grid 134) may be collectively referred to as "perception data." Once the perception engine 110 generates the perception data, the perception engine 110 may provide the perception data to a prediction component and / or a planning component.

[0043]

[0049] The planning component within the computing device 106 may use the radar space grid 134 and / or sensory data, including any determination of drivable and / or non-drivable surfaces as described herein, to locate the autonomous vehicle 102 on a global map and / or local map (which may additionally or alternatively be accomplished by a localization component, not shown), determine one or more trajectories, control the movement of the autonomous vehicle 102 to travel a path or route, and / or otherwise control the operation of the autonomous vehicle 102, although any such operation may be performed in various other components (e.g., localization may be performed by a localization engine, not shown). For example, the planning component may determine a path for the autonomous vehicle 102 from a first location to a second location and may substantially simultaneously generate multiple potential trajectories for controlling the movement of the autonomous vehicle 102 according to a receding horizon technique (e.g., one microsecond, half second) and based at least in part on the radar spatial grid 134 (e.g., to avoid any detected objects and incorporating the likelihood that an object may be obscured by radar noise and not detected by the radar device). For example, the radar spatial grid 134 includes vehicle detection radar data 136 corresponding to the vehicle 120 and a second vehicle detection radar data 138 corresponding to the truck 124. The perception engine 110 may analyze the radar data to determine estimated noise levels associated with positions on the radar spatial grid 134, radar response thresholds and probability thresholds associated with particular object types, and may determine drivable / non-drivable surfaces on the radar spatial grid 134 from which a trajectory for the autonomous vehicle 102 may be generated. FIG. 1 shows an example of such a trajectory, represented as arrows indicating heading, speed, and / or acceleration, but the trajectory itself may include instructions for a PID controller, which may then operate the drive system of the autonomous vehicle 102.

[0044]

[0050] 2 illustrates an example pedestrian RCS distribution 200 and an example pedestrian Doppler distribution 202. In this example, distributions 200 and 202 are associated with pedestrians, however, additional RCS and Doppler distributions may be associated with other object types (e.g., bicycles, cars, trucks, animals, etc.). Distributions 200 and 202 may be generated by the autonomous vehicle 102 and / or may be generated by separate external computing devices and systems and transmitted to the autonomous vehicle 102 for use in determining object type specific radar response thresholds. The data on which distributions 200 and 202 are based may include pedestrian data captured by radar device 108 and / or sensor 104 of a vehicle moving through a physical environment. For example, a vehicle may capture and store log data including radar response data from pedestrians in the environment and additional sensor data (e.g., camera and / or LIDAR data) used to verify that a radar return signal corresponds to a pedestrian.

[0045]

[0051] In this example, the pedestrian RCS distribution 200 may correspond approximately to a normal distribution (or a Gaussian distribution). However, in other examples, the pedestrian RCS distribution, or the RCS distribution of other object types, may correspond to any possible probability distribution function. Individual pedestrian RCS data points in distribution 200 may be based, for example, on the size of the pedestrian, the surface material associated with the pedestrian (e.g., clothing material, color, reflectivity, etc.), the orientation of the pedestrian relative to the radar device, and / or the surface angle of the pedestrian that reflects the radio waves that give rise to the return signal.

[0046]

[0052] In this example, pedestrian Doppler distribution 202 corresponds to a trimodal distribution. The trimodal distribution in this example may represent a relatively large number of pedestrians moving away from the radar device, moving towards the radar device, and stationary relative to the radar device, with walking speeds of approximately 2-4 m / s from the radar device. However, in other examples, distribution 202 may correspond to a uniform distribution, a normal distribution, or any distribution that represents a probability distribution function based on pedestrian Doppler radar measurements. Individual pedestrian Doppler data points in distribution 202 may be based on pedestrian speed and / or any other measurements captured by the Doppler radar device.

[0047]

[0053] 3 illustrates an exemplary radar data analysis including radar return signals corresponding to a number of object detections in an environment, and an estimated noise level. This example also illustrates the distribution 200 described above with reference to FIG. 2, where probability values ​​and / or confidence levels are applied to the distribution to determine radar response thresholds associated with object types. In this example, the pedestrian RCS distribution 200 includes shaded and non-shaded regions and illustrates a pedestrian RCS threshold 302. The RCS threshold 302 may be based on a probability percentile (e.g., the 5th percentile) of the pedestrian RCS values ​​in the distribution 200. A 5% probability may indicate that 95% of the pedestrians detected by the radar device 108 (e.g., the shaded region of the distribution 200) should have RCS values ​​that meet or exceed the RCS threshold 302.

[0048]

[0054] As described above, the perception engine 110 or other components of the autonomous vehicle 102 may select the probability percentile based on a desired safety criterion and / or confidence level that the radar data area does not harbor a pedestrian. Also, the probability value (e.g., 5%) may be modified to increase or decrease the value and adjust the corresponding RCS threshold 302 to improve vehicle safety and / or driving efficiency. For example, lowering the probability from 5% to 2% may result in a lower RCS threshold to ensure that 98% of pedestrians detected by the radar device 108 should have an RCS value that meets or exceeds the lower RCS threshold. This may result in a higher confidence level that the drivable radar data area does not harbor a pedestrian, but may also result in less drivable surface. In contrast, raising the percentile value from 5% to 8% may result in a higher RCS threshold and more drivable surface, but a lower confidence level that the drivable radar data area does not harbor a pedestrian.

[0049]

[0055] While this example describes using the pedestrian RCS distribution 200 and a desired probability or confidence level (e.g., 5%) to determine the RCS threshold 302, in other examples, the pedestrian Doppler distribution 202 may be used in a similar or identical manner to determine the pedestrian Doppler threshold. For example, in the example Doppler distribution 202, a desired probability of 5% may result in a Doppler threshold of -4 m / s, meaning that 95% of observed pedestrians will have Doppler measurements that meet or exceed the -4 m / s threshold. The pedestrian RCS and Doppler distributions 200 and / or 202 may also be used to determine the pedestrian RCS and / or Doppler range instead of or in addition to the RCS and Doppler threshold. For example, a desired confidence level of 95% for the Doppler value may correspond to a Doppler range between -3 m / s and 3 m / s. In various examples, ranges and / or thresholds of RCS and / or Doppler values ​​may be used to determine drivable and non-drivable surfaces.

[0050]

[0056] Additionally, in some examples, if the direction in which a pedestrian (or other object) is likely to be moving can be determined relative to the autonomous vehicle 102, the pedestrian Doppler distribution 202 can be modified. For example, the autonomous vehicle 102 can use various components and functions described herein (e.g., perception components, maps, localization components, etc.) to determine the vehicle's position in the current environment relative to nearby crosswalks, sidewalks, bike paths, one-way or two-way streets, or other features in the environment. The size, shape, angle, and orientation of these environmental features can be used to determine the direction in which a pedestrian is likely to be moving, and the radar response threshold component 112 can use these directions to determine a modified pedestrian Doppler distribution 202 that reflects a more accurate Doppler distribution of a pedestrian moving in the determined direction / angle relative to the vehicle.

[0051]

[0057] The object type radar response thresholds, including the RCS threshold and / or the Doppler threshold, may also be modified upwardly or downwardly during operation of the vehicle to improve safety and / or driving efficiency of the vehicle as needed. For example, a vehicle component may adjust the radar response thresholds used for one or more object types to change the drivable and non-drivable surfaces determined for the vehicle based on the thresholds. For example, if the environment around the vehicle does not contain sufficient drivable surfaces to allow the vehicle to move through the environment, the object type threshold component may increase the radar response threshold to increase the amount of drivable surfaces available to the vehicle. In contrast, if the environment contains more than enough drivable surfaces, the object type threshold component may increase one or more radar response thresholds to increase the confidence level that a potential object was not obscured by radar noise in the environment.

[0052]

[0058] 3 also illustrates a chart 304 showing an example of received radar data. The radar data shown in the chart 304 includes received power levels and distance data associated with received radar data from an example environment. In this example, the chart 304 includes an example object detection 306 and an estimated radar noise level 308 based on the object detection 306. The estimated radar noise level 308 at a location (e.g., distance from the radar device) may be based at least in part on the distance difference between the location and the object detection 306 and the power level of the object detection 306. In some cases, the estimated radar noise level 308 may be based on multiple object detections and / or other attributes of radar data from surrounding or nearby cells within the scanned area. Examples of various techniques for determining an estimated radar noise floor that may be incorporated into an autonomous vehicle 102 to assist in determining the location of potentially hidden objects and determining drivable / non-drivable surfaces can be found, for example, in U.S. patent application Ser. No. 16,407,139, filed May 8, 2019, entitled “Radar False Negative Analysis,” which is incorporated by reference in its entirety herein for all purposes.

[0053]

[0059] The radar data chart 304 also includes two examples of radar return signals 310 and 312 that represent pedestrians in the environment at locations relatively close to the location of the object detection 306. As shown in the chart 304, the radar data associated with the object detection 306 overlaps with the radar data from the radar return signals 310 and 312 in that the radar data from the different detections may affect the same / overlapping areas in the range and / or Doppler dimensions. In this example, the radar return signals 310 and 312 have the same signal power, which may be the same as the RCS threshold 302. The first pedestrian return signal 310 is closer (within range) to the location of the object detection 306 and therefore may not be detected by the perception engine 110 as a separate object detection. In contrast, a second pedestrian return signal 312 having the same power level but farther (in distance) from the location of the object detection 306 exceeds the estimated radar noise level 308 and may therefore be detected as a separate object detection by the perception engine 110. Although various pedestrian return signals may be distributed at any RCS level within the distribution 200, this example shows that the same pedestrian at the RCS threshold 302 may or may not be detected based on the difference in distance between the pedestrian and the location of the object detection 306. As explained below, the point where the RCS threshold 302 crosses the estimated radar noise level 308 may correspond to the boundary of a non-drivable surface surrounding the object detection 306.

[0054]

[0060] FIG. 4 illustrates an example grid map 400 showing the environment around the autonomous vehicle 102. The example grid map 400 may correspond to the image 118 described above with reference to FIG. 1 and may represent the same environment surrounding the autonomous vehicle 102. The grid map 400 also includes grid lines corresponding to regions (e.g., groupings of one or more radar cells) and shaded regions indicating drivable and non-drivable regions in the environment. In this example, two non-drivable regions are shown corresponding to two vehicle radar detections 402 and 404. The vehicle radar detection 402 may correspond to the vehicle detection radar data 136 and the vehicle radar detection 404 may correspond to the second vehicle detection radar data 138. In this example, the boundary between the black (non-drivable) and white (drivable) regions may be where the estimated radar noise level crosses a radar response threshold associated with the object type. For example, if a vehicle radar detection 402 corresponds to an object detection 306 in FIG. 3, the boundary of the non-drivable (blackened) region of the vehicle radar detection 402 may be the location where the estimated radar noise level 308 crosses the pedestrian RCS threshold 302.

[0055]

[0061] As described above, different radar response thresholds, including RCS thresholds and / or Doppler thresholds, may be associated with different object types. Thus, while the exemplary grid map 400 illustrates boundaries between drivable / non-drivable surfaces for the likelihood of a hidden pedestrian in radar data noise, in other examples, different sets of boundaries may be determined for different object types (e.g., bicycles, cars, trucks, animals, etc.). Thus, the perception engine 110 may determine multiple likelihoods for different object types for a single cell. The grid map 400 may have different drivable and non-drivable surfaces associated with a first object type (e.g., “pedestrian”, “large vehicle”, “passenger car”, “cyclist”, “four-legged animal”). The grid map 400 may additionally or alternatively include an indication that individual cells of the grid map 400 corresponding to locations in the environment surrounding the autonomous vehicle 102 are associated with estimated noise levels that do not meet or exceed thresholds specific to various object types, or that are not associated with suitable radar data (e.g., due to occlusion).

[0056]

[0062] FIG. 5 illustrates a graph 500 showing the difference (or delta) in range and Doppler measurements between a detection 502 of a larger object and two different object return signals (first object return signal 504 and second object return signal 506) corresponding to a smaller object in the environment. As discussed above, a larger object in the context of radar data may refer to the magnitude of the radar return signal and not necessarily the physical size of the object. Rather, the magnitude of the radar return signal may depend on a combination of the size, material, orientation, and / or surface angle of the radio wave reflecting object that gives rise to the return signal. In this example, the detection 502 may correspond to a larger object that may generate radar noise that can potentially hide nearby, smaller objects, represented by the first object return signal 504 and the second object return signal 506.

[0057]

[0063] In some examples, at least a portion of the radar noise may correspond to a side lobe level associated with a return signal (e.g., detection 502) of a larger object. The side lobe level associated with an object detection may vary in a variety of different patterns based on the characteristics of the radar device, the radar signal, and the detected object. However, the side lobe level at a nearby location typically decreases in intensity as the location moves further from the location of the object detection. Additionally, in some examples, the side lobe level (and corresponding radar noise) associated with the detection 502 of a larger object may be based on the difference in distance between the detection 502 and the nearby location, and the difference in Doppler measurements between the detection 502 and the nearby location. As shown in this example, the first object return signal 504 has a relatively small difference in distance with the detection 502, but a larger Doppler difference that may increase the amount of radar noise associated with the first object return signal 504. The second object return signal 506 has a relatively large difference in range to the detection 502, as well as a large Doppler difference, which may result in a larger radar noise level associated with the second object return signal 506. Additionally, the Doppler measurement associated with the detection in the radar data may be determined relative to the speed of the autonomous vehicle 102 at the time the Doppler measurement was captured.

[0058]

[0064] 6 is a flow diagram illustrating an example process 600 for analyzing radar data to determine drivable and non-drivable surfaces for a vehicle moving through an environment. As described below, the vehicle may analyze the radar data to determine noise levels in areas near the vehicle based on object detection and may use radar response thresholds associated with particular object types and probability / confidence levels to determine drivable and non-drivable surfaces according to the various systems and techniques described above. In various examples, the operations of process 600 may be performed by one or more components of the autonomous vehicle, such as the radar response threshold component 112, the radar noise estimator 114, the threshold evaluation component 116, and / or the perception engine 110 using various other systems and components described herein.

[0059]

[0065] Process 600 is illustrated as a collection of blocks in a logical flow diagram, which represent a sequence of operations, some or all of which may be implemented in hardware, software, or a combination thereof. In the context of software, the blocks represent computer-executable instructions stored on one or more computer-readable media that, when executed by one or more processors, perform the recited operations. Generally, computer-executable instructions include routines, programs, objects, components, encryption, decryption, compression, recording, data structures, etc. that perform particular functions or implement particular abstract data types. The order in which the operations are described should not be construed as limiting. Any number of the described blocks may be combined in any order and / or in parallel to perform the process or alternative processes, and not all blocks need to be performed. For purposes of explanation, the processes herein are described with reference to the frameworks, architectures, and environments described in the examples herein, although the processes may be implemented in a wide variety of other frameworks, architectures, or environments.

[0060]

[0066] In operation 602, the autonomous vehicle 102 may receive radar data from a radar device indicative of one or more target detections. The radar data may include several return signals received based on objects detected in the environment of the autonomous vehicle 102. The return signals, including radio waves reflected from the objects, may have various characteristics based on characteristics of the detected objects, such as range, azimuth, elevation, Doppler value, received power, SNR, and / or RCS associated with the detected object. In various examples, the radar data received in operation 602 may include either raw radar signal data and / or return signal characteristics derived by the radar device 108 from the raw signal data.

[0061]

[0067] At operation 604, the autonomous vehicle 102 may determine a radar noise level in one or more regions in the environment based on the target detections in the radar data received at operation 602. In some examples, the perception engine 110 may use the radar noise estimator 114 to determine an estimated noise level (or noise floor) of a region in the environment based on characteristics of nearby object detections. As described above with reference to FIG. 3, if a larger object is detected at a detection location in the environment, the detection of the larger object may cause noise in the radar data received for a region near the detection location. For example, for a particular region of the environment, the radar noise estimator 114 may determine an estimated noise level based on the range difference between the particular region and the detection, the Doppler difference between the particular region and the detection, and the strength and / or received power of the detection.

[0062]

[0068] At operation 606, the autonomous vehicle 102 may determine one or more radar response thresholds associated with a particular object type. As described above, the radar response thresholds may include an RCS threshold, a Doppler threshold, and / or other types of radar data thresholds (e.g., intensity or reflected power value thresholds, etc.) associated with the radar data received at operation 602. In some examples, the radar response threshold component 112 may determine the radar response thresholds using one or more distributions associated with a particular object type. For example, based on a predetermined probability or confidence level for detecting a pedestrian (e.g., 95%), the radar response threshold component 112 may use the pedestrian RCS distribution 200 to determine a pedestrian RCS threshold 302 applicable to detect 95% of pedestrians that may be present in the environment. Similar techniques may be used to determine radar response thresholds corresponding to different probability / confidence levels (e.g., 90%, 95%, 99%, etc.), and different distributions (e.g., Doppler distribution 202) may be used to determine different radar response thresholds (e.g., Doppler thresholds and / or distances) for object types. Additionally, as discussed above, radar response threshold component 112 may determine different sets of radar response thresholds for different object types (e.g., pedestrians, bicycles, animals, small vehicles, traffic signs, etc.).

[0063]

[0069] At operation 608, the autonomous vehicle 102 determines whether the radar noise level for a particular region in the environment, determined at operation 604, meets or exceeds the radar response threshold for the object type, determined at operation 606. In some examples, the perception engine 110 may perform several comparisons between the radar noise levels determined by the radar noise estimator 114 for different regions in the environment and the radar response threshold determined by the radar response threshold component 112 for one or more object types. If the estimated radar noise level for a region meets or exceeds the radar response threshold for the object type (608: Yes), then at operation 610, the perception engine 110 may determine that the region is a non-drivable surface. A determination that the region is not drivable at operation 610 may correspond to a determination that the estimated noise level in the region is too high such that the region cannot be effectively scanned by a radar device, providing a desired level of confidence for a particular object type (e.g., 95% of a pedestrian) that no object is hidden in the region. In contrast, if the estimated radar noise level for the region does not exceed the radar response threshold for the object type (608: No), then the perception engine 110 may determine that the region is a drivable surface in operation 612. The determination in operation 612 may correspond to a determination that the estimated noise level in the region is low enough to allow the perception engine 110 to conclude, with a desired confidence level for a particular object type (e.g., 95% for a pedestrian), that no objects are occluded in the region.

[0064]

[0070] In operation 614, the perception engine 110 may determine a trajectory for the autonomous vehicle 102 and / or activate or deactivate additional features (e.g., CAS, remote operation, autonomous driving features, etc.) to control the autonomous vehicle 102 based on the determination of the drivable and non-drivable areas in the environment. In some examples, the perception engine 110 (and / or the prediction or planning component) may determine a planning corridor for navigating the autonomous vehicle using a map of the drivable and non-drivable areas in the environment. For example, the planning corridor may be determined based at least in part on the width and / or perception data of the autonomous vehicle received from the perception engine 110 (e.g., the grid map 400 identifying drivable and non-drivable surfaces). In some examples, the planning corridor may combine potential trajectories generated by the computing device 106 and select a trajectory for operating the autonomous vehicle 102 from among the potential trajectories to exclude non-drivable surfaces and include drivable surfaces in the grid map 400. The planned path may additionally or alternatively be based at least in part on the vehicle width and / or tolerances associated with operating the autonomous vehicle 102.

[0065]

[0071] 7 illustrates a radar data model 700 that stores estimated radar noise levels associated with different Doppler range probabilities in different regions in an environment. In this example, radar data model 700 is a multi-dimensional model that stores multiple estimated radar noise levels for each region in the environment. A region in this example may correspond to one or more radar cells and may be represented by an intersection of azimuth and range values. In other examples, a region may include one or more additional parameters or dimensions (e.g., elevation angle) instead of or in addition to the azimuth and range values.

[0066]

[0072] As described above, the side lobe levels (and corresponding radar noise) generated by the detection of a large object in the environment may be based on the range and Doppler differences between the area of ​​object detection and the area for which the radar noise estimate is being determined. As a result, a single area 702 (defined by the intersection between range and azimuth) in the environment may have different estimated noise levels based on Doppler measurements of objects (e.g., pedestrians) that may potentially be present in the area 702. In some examples, the radar noise estimator 114 may determine different estimated radar noise levels for areas corresponding to different Doppler values. As shown in the radar data model 700, each area in the environment may have multiple different Doppler values ​​(or ranges of Doppler values). For example, for region 702, the model may define multiple Doppler values ​​704, 706, 708, 710, 712, and 714 based on the pedestrian Doppler distribution 202 and may include a distinct radar noise level (e.g., RCS noise) associated with each of the different Doppler values ​​704-714. To illustrate, in region 702, radar noise estimator 114 may determine a first radar noise level value (e.g., between -6 m / s and -4 m / s) associated with Doppler value 704, a second radar noise level value (e.g., between -4 m / s and -2 m / s) associated with Doppler value 706, a third radar noise level value (e.g., between -2 m / s and 0 m / s) associated with Doppler value 708, etc.

[0067]

[0073] In this example, to determine an overall RCS threshold associated with region 702, threshold assessment component 116 may use the individual radar noise level values ​​associated with the different Doppler values ​​704-714 (e.g., based on pedestrian Doppler distribution 202) along with the probabilities associated with the different Doppler values ​​704-714. For example, the probability function used to determine the RCS threshold for region 702 based on the individual RCS thresholds associated with the different Doppler values ​​corresponds to a 95% confidence level that no pedestrians are occupying the region, expressed as follows according to Equation 1:

[0068]

Number

[0069] In this example, each P(vel = N m / s) represents the probability that a pedestrian moves at N m / s in the environment, and each P(RCS < est RCS noise level at -N m / s) represents the probability that a particular RCS is less than the estimated radar noise level value (e.g., N m / s) associated with the same Doppler measurement. Using Equation 1, the threshold evaluation component 116 can solve for the RCS threshold corresponding to the desired probability / reliability level (e.g., 95%) at which a pedestrian cannot hide in the area. Also, Equation 1 may be executed using a predetermined RCS term to solve for the overall probability (which may be greater than or less than 0.95 in this example) associated with the same predetermined RCS threshold.

[0070]

[0074] FIG. 8 is a flowchart showing another exemplary process 800 for analyzing radar data to determine drivable and non-drivable surfaces of a vehicle moving in an environment. In some examples, process 800 may be the same as process 600 described above. However, in this example, a single region may have multiple different estimated radar noise levels associated with different Doppler values (or ranges of Doppler values), and the vehicle may determine individual probabilities associated with the different Doppler values and sum the individual probabilities to determine the overall probability associated with the region. As described below, the operations of process 800 may be performed by one or more components of the autonomous vehicle 102, such as the radar response threshold component 112, the radar noise estimator 114, the threshold evaluation component 116, and / or the perception engine 110 including various other systems and components described herein.

[0071]

[0075] At operation 802, the autonomous vehicle 102 may receive radar data from a radar device indicative of one or more target detections. In some examples, operation 802 may be similar or identical to operation 602 described above, where the radar data may include one or more return signals received based on objects detected in the environment of the autonomous vehicle 102.

[0072]

[0076] In operation 804, the autonomous vehicle 102 may determine an area within the environment, and the radar data is analyzed to determine whether the area is a drivable or non-drivable surface. The area may be defined by the intersection of a range value and an azimuth value (and / or an elevation value) relative to the radar device 108. In some examples, the determined area may be a vicinity of one or more radar target detections received in operation 802.

[0073]

[0077] In operation 806, the autonomous vehicle 102 may determine a set of Doppler probabilities for an object type (e.g., a pedestrian) and corresponding estimated radar noise levels associated with the Doppler probabilities. For example, as described above with reference to FIG. 7, pedestrians or other object types may be associated with different ranges of Doppler values ​​(e.g., −5 m / s to 5 m / s) and may have different probabilities associated with each Doppler value. These probabilities may be determined based on the pedestrian Doppler distribution 202 for pedestrians or for different Doppler distributions associated with different object types. Additionally, because each different Doppler value may have a different associated radar noise level, the threshold evaluation component 116 may perform different probability evaluations using different RSC thresholds for each different Doppler value.

[0074]

[0078] In operation 808, the threshold evaluation component 116 may determine different Doppler values ​​(or ranges of Doppler values) associated with the object type and the estimated radar noise level associated with each Doppler value in the region. In operation 810, the threshold evaluation component 116 may evaluate the probability that an object of a particular object type (e.g., a pedestrian) will be obscured by the estimated radar noise in a region of a particular Doppler value of the object. For example, the probability determination in operation 810 may represent the probability that a pedestrian moving at N m / s will be obscured by the radar noise associated with the N m / s Doppler value. Thus, in operation 810, the threshold evaluation component 116 may use the pedestrian Doppler distribution 202 to determine the probability that a pedestrian present in the region will be obscured by the radar noise level associated with a Doppler value of N m / s (e.g., corresponding to the percentage of pedestrians having an RCS less than the radar noise level) and multiply that probability by the probability that a pedestrian present in the region moves at a Doppler value of N m / s. As shown in this example, and as shown in equation 1 above, the determination in operation 810 may be performed iteratively for each different Doppler value at which an object (e.g., a pedestrian) may potentially be present in the area.

[0075]

[0079] In operation 812, the threshold evaluation component 116 may sum the individual probabilities that a pedestrian would be obscured by radar noise levels at a particular Doppler value across all of the different Doppler values ​​associated with the object to determine a total probability that a pedestrian would be obscured by radar noise in that area. Equation 1 provides an example of a probability sum that may be similar or identical to the sum in operation 812.

[0076]

[0080] In operation 814, the autonomous vehicle 102 determines whether the total probability determined in operation 812 (e.g., the probability that an object will be obscured by radar noise in the region) is greater than a predetermined probability / confidence level associated with the object type. For example, a predetermined probability of 5% may represent a 95% confidence level that a pedestrian in the region will be detected in the radar data. In some examples, the comparison in operation 814 may be similar or identical to the comparison in operation 608 described above. In this example, if the probability that a pedestrian present in the region will be obscured by radar noise meets or exceeds the predetermined probability for a pedestrian (814: Yes), then in operation 816, the perception engine 110 may determine that the region is a non-drivable surface. As described above, a determination that the region is not drivable in operation 816 may correspond to a determination that the estimated noise level in the region (e.g., across different Doppler values / ranges) is too high to allow the region to be effectively scanned by a radar device to ensure a desired level of confidence for a particular object type (e.g., 95% for a pedestrian) that no object is hidden in the region. In contrast, if the probability that a pedestrian present in the region would be hidden by radar noise does not exceed a predetermined probability for a pedestrian (814: No), then in operation 818 the perception engine 110 may determine that the region is a drivable surface. The determination in operation 818 may correspond to a determination that the estimated noise level in the region (e.g., across different Doppler values) is sufficiently low to allow the perception engine 110 to conclude that no object is hidden in the region, with a desired level of confidence for a particular object type (e.g., 95% for a pedestrian).

[0077]

[0081] In some examples, during or after operations 816 and 818, the perception engine 110 may determine a trajectory of the autonomous vehicle 102 and / or activate or deactivate additional features (e.g., CAS, remote operation, autonomous driving functions, etc.) to control the autonomous vehicle 102. For example, the perception engine 110 may control the autonomous vehicle based on the determination of drivable and non-drivable areas in the environment using techniques similar or identical to those of operation 614. Additionally, while the particular example above is described with respect to determining the probability that a pedestrian will be obscured by radar noise levels in an area, in other examples, similar or identical techniques may be used to determine the probability of different object types. For example, the perception engine 110 may determine different grid maps having different drivable and non-drivable surfaces defined for different object types.

[0078]

[0082] FIG. 9 illustrates a block diagram of an example system 900 for implementing the techniques described herein. In at least one example, the system 900 may include a vehicle 902, which may correspond to an autonomous or semi-autonomous vehicle configured to perform object perception and prediction functions, path planning and / or optimization. As described below, the vehicle 902 may include components configured to analyze radar detections to detect objects potentially hidden within the radar noise. For example, the vehicle 902 may be similar or identical to the autonomous vehicle 102 described above and may include components similar or identical to the perception engine 110, the prediction and / or planning component, the radar response threshold component 112, the radar noise estimator 114 and / or the threshold evaluation component 116. The exemplary vehicle 902 may be a driverless vehicle, such as an autonomous vehicle configured to operate in accordance with a Level 5 classification issued by the National Highway Traffic Safety Administration, which describes a vehicle capable of performing all safety-critical functions for the entire trip in situations where a driver (or passenger) is not expected to control the vehicle at any time. In such an example, the vehicle 902 may not include a driver and / or controls for driving the vehicle 902, such as a steering wheel, accelerator pedal, and / or brake pedal, since the vehicle 902 may be configured to control all functions from start to finish of a trip, including all parking functions. This is merely an example, and the systems and methods described herein may be incorporated into any land-based, air-based, or water-based vehicle, including vehicles that require manual control by a driver at all times, to vehicles that are partially or fully autonomously controlled.

[0079]

[0083] In this example, vehicle 902 may include a vehicle computing device 904, one or more sensor systems 906, one or more emitters 908, one or more communication connections 910, at least one direct connection 912, and one or more drive systems 914.

[0080]

[0084] The vehicle computing device 904 may include one or more processors 916 and a memory 918 communicatively coupled to the one or more processors 916. In the illustrated example, the vehicle 902 is an autonomous vehicle, although the vehicle 902 may be other types of vehicles or robotic platforms. In the illustrated example, the memory 918 of the vehicle computing device 904 stores a localization component 920, a perception component 922, a radar response threshold component 924, a radar noise estimator 925, a threshold evaluation component 926, one or more maps 928, one or more system controllers 930, a prediction component 932, and a planning component 934. While shown in FIG. 9 as residing in memory 918 for illustrative purposes, one or more of the localization component 920, perception component 922, radar response threshold component 924, radar noise estimator 925, threshold evaluation component 926, map 928, system controller 930, prediction component 932, and planning component 934 may additionally or alternatively be accessible to the vehicle 902 (e.g., stored in memory separate from the vehicle 902 or otherwise accessible by the vehicle 902).

[0081]

[0085] In at least one example, the localization component 920 can include functionality for receiving data from the sensor system 906 to determine a position and / or orientation (e.g., one or more of x-, y-, z-position, roll, pitch, or yaw) of the vehicle 902. For example, the localization component 920 can include and / or request / receive a map of the environment and can continually determine the position and / or orientation of the autonomous vehicle within the map. In some examples, the localization component 920 can receive image data, lidar data, radar data, time-of-flight data, IMU data, GPS data, wheel encoder data, etc., to precisely determine the position of the autonomous vehicle using SLAM (simultaneous localization and mapping), CLAMS (simultaneous calibration, localization, and mapping), relative SLAM, bundle adjustment, nonlinear least squares optimization, etc. In some examples, the localization component 920 can provide data to various components of the vehicle 902 to determine an initial position of the autonomous vehicle, generate a trajectory, and / or determine that an object is in proximity to one or more pedestrian crossing areas and / or identify candidate reference lines, as discussed herein.

[0082]

[0086] In some examples, the perception component 922 may include functionality for performing object detection, segmentation, and / or classification. In some examples, the perception component 922 may be similar or identical to the perception engine 110 and may provide processed sensor data indicative of the presence of an entity in proximity to the vehicle 902 and / or a classification of the entity as a type of entity (e.g., car, pedestrian, bicycle, animal, building, tree, road surface, curb, sidewalk, stop light, stop sign, unknown, etc.). In additional or alternative examples, the perception component 922 may provide processed sensor data indicative of one or more characteristics associated with a detected entity (e.g., a tracked object) and / or an environment in which the entity is located. In some examples, characteristics associated with an entity may include, but are not limited to, x position (global and / or local position), y position (global and / or local position), z position (global and / or local position), orientation (e.g., roll, pitch, yaw), entity type (e.g., classification), entity velocity, entity acceleration, entity range (size), etc. Characteristics associated with an environment may include, but are not limited to, the presence of other entities in the environment, the state of other entities in the environment, time of day, day of the week, season, weather conditions, darkness / lightness indicators, etc.

[0083]

[0087] As shown in this example, the perception component 922 can include a radar response threshold component 924, a radar noise estimator 925, and / or a threshold evaluation component 926. The radar response threshold component 924, the radar noise estimator 925, and the threshold evaluation component 926 may perform similar or identical functions as the radar response threshold component 112, the radar noise estimator 114, and the threshold evaluation component 116 described above. For example, the radar response threshold component 924 may be configured to determine a radar response threshold that may be applied by the autonomous vehicle 902 while operating in a driving environment. The radar noise estimator 925 may be configured to determine an estimate of radar noise (e.g., RCS noise and / or Doppler noise) based on detection of objects by one or more radar devices on the vehicle 902. As described above, the radar noise may be based on a sidelobe level received in response to a radar transmission signal. The radar noise at a location near the target detection may be based on the range difference between the target detection and the location, the Doppler difference between the target detection and the location, and the power (e.g., intensity) of the target detection. The threshold evaluation component 926 may determine and / or store radar response thresholds (e.g., RCS and / or Doppler thresholds) associated with a particular type of object (e.g., pedestrian, bicycle, animal, car, etc.). As described above, object type-specific thresholds of radar response may be based on object-specific RCS and Doppler distributions and may be adjusted based on a desired probability that a location does not contain an obscured object of the object type. The perception component 922 may include functionality to modify and adjust thresholds and compare the thresholds to estimated radar noise at various locations to determine drivable and / or non-drivable surfaces around the vehicle 902. As shown in this example, the radar response threshold component 924, the radar noise estimator 925, and / or the threshold evaluation component 926 may be implemented within the perception component 922.However, in other examples, one or more of the radar response threshold component 924, the radar noise estimator 925, and / or the threshold evaluation component 926 may be implemented within the prediction component 932, within the planning component 934, or elsewhere within the vehicle computing device 904.

[0084]

[0088] The memory 918 may further include one or more maps 928 that may be used by the vehicle 902 to navigate within the environment. For purposes of this disclosure, a map may be any number of data structures modeled in two, three, or N dimensions that may provide information about the environment, such as topology (such as intersections), roads, mountain ranges, roads, terrain, and the environment in general, but is not limited to such. In some examples, the map may include, but is not limited to, texture information (e.g., color information (e.g., RGB color information, Lab color information, HSV / HSL color information), etc.), intensity information (e.g., lidar information, radar information, etc.), spatial information (e.g., vectorized information about features of the environment, image data projected onto a mesh, individual "surfels" (e.g., polygons associated with individual colors and / or intensities), reflectance information (e.g., specular reflectance information, retroreflectance information, BRDF information, BSSRDF information, etc.). In one example, the map may include a three-dimensional mesh of the environment. In some examples, the maps may be stored in a tiled format, with each tile of the map representing a discrete portion of the environment, and may be loaded into the working memory as needed. In at least one example, the one or more maps 928 may include at least one map (e.g., an image and / or a mesh).

[0085]

[0089] In some examples, vehicle 902 may be controlled based at least in part on map 928. That is, map 928 may be used in conjunction with localization component 920, perception component 922, prediction component 932, and / or planning component 934 to determine a position of vehicle 902, identify objects within the environment, and / or generate a path and / or trajectory for navigating within the environment.

[0086]

[0090] In some examples, one or more maps 928 may be stored on a remote computing device, such as in memory 942 of computing device 938, and may be accessible to vehicle 902 via network 936. In some examples, multiple maps 928 may be retrieved from memory 942 and stored, for example, based on characteristics (e.g., type of entity, time of day, day of the week, season, etc.). Storing multiple maps 928 may have similar memory requirements but may increase the speed at which data in the maps can be accessed.

[0087]

[0091] In at least one example, vehicle computing device 904 may include one or more system controllers 930, which may be configured to control steering, propulsion, braking, safety, emitter, communication, and other systems of vehicle 902. These system controllers 930 may communicate with and / or control corresponding systems of drive system 914 and / or other components of vehicle 902. For example, planning component 934 may generate instructions based at least in part on perception data generated by perception component 922 (which may include any of the radar space grids and / or likelihoods discussed herein) and transmit the instructions to system controller 930, which may control operation of vehicle 902 based at least in part on the instructions. In some examples, if the planning component 934 receives notification that tracking of an object has been “lost” (e.g., the object appears in the LIDAR but is no longer visible in the sensor data that is not occluded by other objects), the planning component 934 may generate instructions to bring the vehicle 902 to a safety stop and / or send a request for remote operation assistance.

[0088]

[0092] In general, the prediction component 932 may include functionality for generating predictive information associated with objects in an environment. As an example, the prediction component 932 may be implemented to predict the location of a pedestrian proximate a crosswalk area (or an area or location otherwise associated with a pedestrian crossing a road) in an environment as the pedestrian crosses the crosswalk area or prepares to cross the crosswalk area. As another example, the techniques discussed herein may be implemented to predict the location of other objects (e.g., vehicles, bicycles, pedestrians, etc.) as the vehicle 902 moves through an environment. In some examples, the prediction component 932 may generate one or more predicted positions, predicted speeds, predicted trajectories, etc. for target objects based on attributes of such target objects and / or other objects proximate the target objects.

[0089]

[0093] In general, the planning component 934 can determine a path for the vehicle 902 to follow to move through an environment. The planning component 934 can determine various routes and trajectories and various levels of detail. For example, the planning component 934 can determine a route to travel from a first location (e.g., a current location) to a second location (e.g., a target location). For purposes of this discussion, the route can be a sequence of waypoints for traveling between the two locations. As non-limiting examples, the waypoints include streets, intersections, Global Positioning System (GPS) coordinates, and the like. Additionally, the planning component 934 can generate instructions for guiding the autonomous vehicle along at least a portion of the route from the first location to the second location. In at least one example, the planning component 934 can determine how to guide the autonomous vehicle from a first waypoint in the sequence of waypoints to a second waypoint in the sequence of waypoints. In some examples, the instructions can be a trajectory or a portion of a trajectory. In some examples, multiple trajectories can be generated substantially simultaneously (e.g., within technical tolerances) according to a receding horizon technique, and one of the multiple trajectories is selected for navigating the vehicle 902.

[0090]

[0094] In some examples, the planning component 934 can generate one or more trajectories for the vehicle 902 based at least in part on predicted positions associated with objects in the environment. In some examples, the planning component 934 can evaluate one or more trajectories for the vehicle 902 using temporal logic, such as linear temporal logic and / or signal temporal logic.

[0091]

[0095] As can be understood, the components discussed herein (e.g., localization component 920, perception component 922, one or more maps 928, one or more system controllers 930, prediction component 932, and planning component 934) are described as separated for illustrative purposes. However, operations performed by the various components may be combined or performed in any component. Additionally, any component described as implemented in software may be implemented in hardware and vice versa. Additionally, any functionality implemented in vehicle 902 may be implemented in computing device 938 or other components (and vice versa).

[0092]

[0096] In at least one example, the sensor system 906 may include time-of-flight sensors, lidar sensors, radar devices and / or sensors, ultrasonic transducers, sonar sensors, position sensors (e.g., GPS, compass, etc.), inertial sensors (e.g., inertial measurement units (IMUs), accelerometers, magnetometers, gyroscopes, etc.), cameras (RGB, IR, intensity, depth, etc.), microphones, wheel encoders, environmental sensors (temperature sensors, humidity sensors, light sensors, pressure sensors, etc.), and the like. The sensor system 906 may include multiple instances of each of these or other types of sensors. For example, the time-of-flight sensors may include individual time-of-flight sensors located at the corners, front, back, sides, and / or top of the vehicle 902. As another example, the camera sensors may include multiple cameras located at various locations about the exterior and / or interior of the vehicle 902. The sensor system 906 may provide input to the vehicle computing device 904. Additionally or alternatively, the sensor system 906 may transmit sensor data via one or more networks 936 to one or more computing devices 938 at a particular frequency, after a predetermined period of time, in near real-time, etc.

[0093]

[0097] The vehicle 902 may also include one or more emitters 908 that emit light and / or sound, as described above. The emitters 908, in this example, include interior audio and visual emitters for communicating with passengers of the vehicle 902. By way of example and not by way of limitation, the interior emitters may include speakers, lights, signs, display screens, touch screens, haptic emitters (e.g., vibration and / or force feedback), mechanical actuators (e.g., seat belt tensioners, seat positioners, head rest positioners, etc.), and the like. The emitters 908, in this example, may also include exterior emitters. By way of example and not by way of limitation, the exterior emitters in this example may include lights that indicate direction of travel or other indicators of the vehicle's actions (e.g., indicator lights, signs, light arrays, etc.), as well as one or more audio emitters (e.g., speakers, speaker arrays, horns, etc.) for acoustically communicating with pedestrians or other nearby vehicles, one or more of which may include acoustic beam steering technology.

[0094]

[0098] Vehicle 902 may also include one or more communication connections 910 that enable communication between vehicle 902 and one or more other local or remote computing devices. For example, communication connections 910 may facilitate communication with other local computing devices on vehicle 902 and / or drive system 914. Communication connections 910 may also allow the vehicle to communicate with other nearby computing devices (e.g., other nearby vehicles, traffic signals, etc.). Communication connections 910 also enable vehicle 902 to communicate with remotely operated computing devices or other remote services.

[0095]

[0099] The communication connections 910 may include physical and / or logical interfaces for connecting the vehicle computing device 904 to other computing devices or networks, such as network 936. For example, the communication connections 910 may enable Wi-Fi®-based communications, such as over frequencies defined by the IEEE 802.11 standard, short-range wireless frequencies such as Bluetooth®, cellular communications (e.g., 2G, 3G, 4G, 4G LTE, 5G, etc.), or any suitable wired or wireless communication protocol that allows each computing device to interface with other computing devices.

[0096]

[0100] In at least one example, the vehicle 902 may include one or more drive systems 914. The vehicle 902 may have a single drive system 914 or may have multiple drive systems 914. In at least one example, when the vehicle 902 has multiple drive systems 914, the individual drive systems 914 may be located at opposite ends of the vehicle 902 (e.g., the front and rear, etc.). In at least one example, the drive system 914 may include one or more sensor systems for detecting conditions surrounding the drive system 914 and / or the vehicle 902. By way of example and not by way of limitation, the sensor systems may include one or more wheel encoders (e.g., rotary encoders) for sensing the rotation of the wheels of the drive module, inertial sensors (e.g., inertial measurement units, accelerometers, gyroscopes, magnetometers, etc.) for measuring the orientation and acceleration of the drive module, cameras or other image sensors, ultrasonic sensors for acoustically detecting objects surrounding the drive system, lidar sensors, radar sensors, etc. Some sensors, such as the wheel encoders, may be unique to the drive system 914. In some cases, sensor systems on drive system 914 may overlap or complement corresponding systems on vehicle 902 (eg, sensor system 906).

[0097]

[0101] The drive system 914 can include a high voltage battery, a motor to propel the vehicle, an inverter to convert direct current from the battery to alternating current for use by other vehicle systems, a steering system including a steering motor and a steering rack (which can be electric), a braking system including hydraulic or electric actuators, a suspension system including hydraulic and / or pneumatic components, a stability control system to distribute braking force to mitigate loss of traction and maintain control, an HVAC system, lights (e.g., lights such as head / tail lights that illuminate the exterior surroundings of the vehicle), and one or more other systems (e.g., cooling systems, safety systems, on-board charging systems, DC / DC converters, high voltage junctions, high voltage cables, charging systems, charging ports, and other electrical components). Additionally, the drive system 914 can include a drive system controller that can receive and preprocess data from the sensor systems to control the operation of various vehicle systems. In some examples, the drive system controller can include one or more processors and a memory communicatively coupled to the one or more processors. The memory can store one or more components for performing various functions of the drive system 914. Additionally, drive systems 914 may also include one or more communication connections that enable each drive system to communicate with one or more other local or remote computing devices.

[0098]

[0102] In at least one example, the direct connection 912 can provide a physical interface for coupling one or more drive systems 914 with the body of the vehicle 902. For example, the direct connection 912 can allow for the transfer of energy, fluid, air, data, etc. between the drive systems 914 and the vehicle. In some examples, the direct connection 912 can releasably secure the drive systems 914 to the body of the vehicle 902.

[0099]

[0103] In at least one example, the localization component 920, the perception component 922, the radar response threshold component 924, the radar noise estimator 925, the threshold evaluation component 926, the one or more maps 928, the one or more system controllers 930, the prediction component 932, and the planning component 934 may process the sensor data as described above and may transmit their respective outputs to one or more computing devices 938 via one or more networks 936. In at least one example, the output of each of the components may be transmitted to the one or more computing devices 938 at a particular frequency, after a predetermined period of time, in near real time, etc. Additionally or alternatively, the vehicle 902 may transmit sensor data, including raw sensor data, processed sensor data, and / or representations of sensor data, to the one or more computing devices 938 via the network 936. Such sensor data may be transmitted to the computing device 938 as one or more log files at a particular frequency, after a predetermined period of time, in near real time, etc.

[0100]

[0104] The computing device 938 may include a processor 940 and a memory 942 that stores one or more radar-responsive object profiles 944 and / or vehicle safety metrics 946. As described above, the radar-responsive object profiles 944 may include response data (e.g., RSC data and / or Doppler data) associated with a variety of different object types (e.g., pedestrians, bicycles, animals, cars, etc.). The radar-responsive object profiles 944 may include individual values, distributions, and / or probability or confidence metrics associated with the different object types. The vehicle safety metrics 946 may include crash and safety metrics such as miles per crash, injury or fatality estimates. In various examples, the computing device 938 may implement one or more heuristic-based systems and / or neural network models to determine the radar-responsive object profiles 944 based on the vehicle safety metrics 946 and / or log data received from the vehicle 902 and / or additional vehicles operating in the environment. Additionally, any of the features or functions described in connection with the radar response threshold component 924, the radar noise estimator 925 (e.g., determining an estimated radar noise based on radar detection), and / or the threshold evaluation component 926 (e.g., determining object type specific probabilities and thresholds) may be implemented using heuristics-based techniques and / or neural network models and algorithms. In this example, a neural network is an algorithm that passes input data through a series of connected layers to generate an output. Individual layers of a neural network may constitute other neural networks, and may constitute any number of layers (convolutional or not). As can be understood in the context of this disclosure, a neural network may utilize machine learning, which may refer to a broad class of algorithms in which an output is generated based on learned parameters. Any type of machine learning may be used consistent with this disclosure.

[0101]

[0105] The processor 916 of the vehicle 902 and the processor 940 of the computing device 938 may be any suitable processor capable of processing data and executing instructions to perform operations as described herein. By way of example and not limitation, the processors 916 and 940 may include one or more central processing units (CPUs), graphics processing units (GPUs), or any other device or portion of a device that processes electronic data and converts it into other electronic data that may be stored in registers and / or memory. In some examples, integrated circuits (e.g., ASICs, etc.), gate arrays (e.g., FPGAs, etc.), and other hardware devices may also be considered processors so long as they are configured to implement encoded instructions.

[0102]

[0106] The memories 918 and 942 are examples of non-transitory computer-readable media. The memories 918 and 942 may store an operating system and one or more software applications, instructions, programs, and / or data for implementing the functions resulting from the methods and various systems described herein. In various implementations, the memories may be implemented using any suitable memory technology, such as static random access memory (SRAM), synchronous dynamic RAM (SDRAM), non-volatile / flash memory, or any other type of memory capable of storing information. The architectures, systems, and individual elements described herein may include many other logical, programmatic, and physical components, of which the ones shown in the accompanying drawings are merely examples relevant to the discussion herein.

[0103]

[0107] 9 is shown as a distributed system, it should be noted that in alternative examples, components of the vehicle 902 may be associated with the computing device 938 and / or components of the computing device 938 may be associated with the vehicle 902. That is, the vehicle 902 may perform one or more functions associated with the computing device 938, or vice versa.

[0104] Example clause

[0108] A. A system comprising one or more processors and one or more non-transitory computer-readable media storing computer-executable instructions that, when executed, cause the system to perform operations including receiving radar data from a radar device associated with a vehicle operating within an environment, the radar data indicating a detection associated with a first location within the environment, determining a radar noise level associated with a second location within the environment based at least in part on the radar data, determining a vehicle safety metric associated with the vehicle, determining a radar response threshold based at least in part on the vehicle safety metric, and generating a trajectory for the vehicle based at least in part on comparing the radar response threshold to the radar noise level associated with the second location.

[0105]

[0109] B. The system of paragraph A, wherein the vehicle safety metric is associated with an object type and determining the radar response threshold is based at least in part on a radar response distribution associated with the object type.

[0106]

[0110] C. The system of paragraph A, wherein determining the radar response threshold includes determining a first number of false positive detections associated with the radar response threshold, determining a second number of false negative detections associated with the radar response threshold, determining a third number of true positive detections associated with the radar response threshold, and determining a third number of true negative detections associated with the radar response threshold.

[0107]

[0111] D. The system of paragraph A, wherein the operations further include determining a second radar noise level associated with a third location within the environment based at least in part on the radar data, and generating a trajectory for the vehicle further includes generating a trajectory that excludes the second location and includes the third location, based at least in part on determining that the radar noise level associated with the second location meets or exceeds a radar response threshold and determining that the second radar noise level associated with the third location is below a radar response threshold.

[0108]

[0112] E. The system of paragraph A, wherein the operations further include determining a drivable surface metric associated with the vehicle, and wherein determining the radar response threshold is based at least in part on the drivable surface metric.

[0109]

[0113] F. A method including receiving radar data associated with an area within an environment, determining radar noise data associated with the area, evaluating the radar data based at least in part on a radar response threshold for a vehicle and the radar noise data, the radar response threshold being based at least in part on a vehicle safety metric, and controlling a vehicle in the environment based at least in part on the radar response threshold and the radar data.

[0110]

[0114] G. The method of paragraph F, wherein the vehicle safety metric is associated with a first object type, the method further including determining a probability associated with detecting an object of the first object type at a first location in the environment based at least in part on the radar noise data and a radar response distribution associated with the first object type, and determining the radar response threshold based at least in part on the probability.

[0111]

[0115] H. The method of paragraph G, further comprising: determining a second radar response threshold associated with a second object type based at least in part on the radar noise data and a second radar response distribution associated with the second object type; and determining a second probability associated with detecting a second object of the second object type at the first location based at least in part on the radar noise data and the second radar response distribution.

[0112]

[0116] I. The method of paragraph F, further including determining a first radar response threshold, determining a first number of false positive radar detections and a first number of false negative radar detections associated with the first radar response threshold, determining a second radar response threshold, determining a second number of false positive radar detections and a second number of false negative radar detections associated with the second radar response threshold, and determining either the first radar response threshold or the second radar response threshold as the radar response threshold based at least in part on the first number of false positive radar detections and the second number of false positive radar detections and the first number of false negative radar detections and the second number of false negative radar detections.

[0113]

[0117] J. The method of paragraph F, further comprising: determining a drivable surface metric associated with operating the vehicle within the environment; and determining the radar response threshold based at least in part on the drivable surface metric.

[0114]

[0118] K. The method of paragraph F, further comprising: determining a first radar noise level associated with a first location within the environment and a second radar noise level associated with a second location within the environment based at least in part on the radar noise data; and generating a trajectory for the vehicle that excludes the first location and includes the second location based at least in part on determining that the first radar noise level meets or exceeds the radar response threshold and determining that the second radar noise level is below the radar response threshold.

[0115]

[0119] L. The method of paragraph F, wherein determining the radar noise data includes determining a first location in the environment associated with the radar data and determining a radar noise level associated with a second location in the environment, based at least in part on a difference in distance from a radar device between the first location and the second location and an intensity measurement associated with the radar data.

[0116]

[0120] M. The method of paragraph L, further comprising: receiving the radar data from the radar device associated with the vehicle; receiving additional sensor data associated with the second location from a sensor device having a different sensor means than the radar device; and determining a trajectory of the vehicle based at least in part on a comparison of the radar response threshold to the radar noise level associated with the second location and the additional sensor data associated with the second location.

[0117]

[0121] N. The method of paragraph F, further comprising: determining a first vehicle safety metric associated with a first environment; determining a second vehicle safety metric associated with a second environment; determining an operating environment of the vehicle; and determining either the first vehicle safety metric or the second vehicle safety metric as the vehicle safety metric based at least in part on the operating environment of the vehicle.

[0118]

[0122] O. One or more non-transitory computer-readable media storing processor-executable instructions that, when executed by one or more processors, cause the one or more processors to perform operations including receiving radar data associated with an area within an environment; determining radar noise data associated with the area; evaluating the radar data based at least in part on a radar response threshold for the vehicle and the radar noise data, the radar response threshold based at least in part on a vehicle safety metric; and controlling a vehicle in the environment based at least in part on the radar response threshold and the radar data.

[0119]

[0123] P. The one or more non-transitory computer-readable media of paragraph O, wherein the vehicle safety metric is associated with a first object type, and the operations further include determining a probability associated with detecting an object of the first object type at a first location in the environment based at least in part on the radar noise data and a radar response distribution associated with the first object type, and determining the radar response threshold based at least in part on the probability.

[0120]

[0124] Q. The one or more non-transitory computer-readable media of paragraph P, wherein the operations further include determining a second radar response threshold associated with a second object type based at least in part on the radar noise data and a second radar response distribution associated with the second object type, and determining a second probability associated with detecting a second object of the second object type at the first location based at least in part on the radar noise data and the second radar response distribution.

[0121]

[0125] R. The one or more non-transitory computer-readable media of paragraph O, wherein the operations further include determining a first radar response threshold; determining a first number of false positive radar detections and a first number of false negative radar detections associated with the first radar response threshold; determining a second radar response threshold; determining a second number of false positive radar detections and a second number of false negative radar detections associated with the second radar response threshold; and determining either the first radar response threshold or the second radar response threshold as the radar response threshold based at least in part on the first number of false positive radar detections and the second number of false positive radar detections and the first number of false negative radar detections and the second number of false negative radar detections.

[0122]

[0126] S. The one or more non-transitory computer-readable media of paragraph O, wherein the operations further include determining a drivable surface metric associated with operating the vehicle within the environment and determining the radar response threshold based at least in part on the drivable surface metric.

[0123]

[0127] T. The one or more non-transitory computer-readable media of paragraph O, wherein the operations further include determining a first radar noise level associated with a first location within the environment and a second radar noise level associated with a second location within the environment based at least in part on the radar noise data, and generating a trajectory for the vehicle that excludes the first location and includes the second location based at least in part on determining that the first radar noise level meets or exceeds the radar response threshold and determining that the second radar noise level is below the radar response threshold.

[0124]

[0128] U. A system comprising one or more processors and one or more non-transitory computer readable media storing computer executable instructions, the instructions, when executed, causing the system to: receive first radar data from a radar device, the first radar data indicative of a detection associated with a first object in an environment, the first radar data including Doppler data associated with the first object and radar return power data associated with the first object; and receive second radar data from the radar device, the second radar data indicative of a detection associated with a second possible object in the environment, the second radar data including Doppler data associated with the second possible object. 16. A system for controlling a vehicle in the environment based at least in part on the probability associated with the second radar data. 17. A system for controlling a vehicle in the environment based at least in part on the probability associated with the second radar data. 18. A system for controlling a vehicle in the environment based at least in part on the probability associated with the second radar data.

[0125]

[0129] V. The system described in paragraph U, wherein controlling the vehicle includes generating a trajectory that excludes a location associated with the second radar data based at least in part on determining that a radar noise level associated with the second radar data meets or exceeds a radar response threshold associated with the object type.

[0126]

[0130] W. The system of paragraph U, wherein the operations further include determining a first radar noise level and a second radar noise level associated with the second radar data, wherein the first radar noise level is a reflected power value and the second radar noise level is a Doppler value; determining a first radar response threshold based at least in part on a reflected power distribution associated with the object type; and determining a second radar response threshold based at least in part on a Doppler distribution associated with the object type; and controlling the vehicle is based at least in part on comparing the first radar noise level to the first radar response threshold and comparing the second radar noise level to the second radar response threshold.

[0127]

[0131] X. The system described in paragraph U, wherein determining the probability associated with the second radar data includes determining a first difference in distance between the first radar data and the second radar data, determining a second difference in Doppler measurements between the first radar data and the second radar data, determining an intensity measurement associated with the first radar data, and determining a radar noise level associated with the second radar data based at least in part on the first difference, the second difference, and the intensity measurement.

[0128]

[0132] Y. The system of paragraph U, wherein determining the probability associated with the second radar data includes determining a range value and an azimuth angle value associated with the second radar data, and determining a first reflected power threshold and a second reflected power threshold based at least in part on the range value and the azimuth angle value, wherein the first reflected power threshold is associated with a first Doppler value and the second reflected power threshold is associated with a second Doppler value.

[0129]

[0133] Z. A method comprising: receiving radar data indicative of a detection associated with a first region within an environment, the radar data including Doppler data associated with the detection; determining a radar noise level associated with a second region based at least in part on the Doppler data associated with the detection; determining a radar response threshold associated with an object type; and controlling a vehicle in the environment based at least in part on the radar noise level associated with the second region and the radar response threshold associated with the object type.

[0130]

[0134] AA. The method of paragraph Z, wherein the radar response threshold includes at least one of a Doppler value associated with the object type, a radar cross section value associated with the object type, or a reflected power value associated with the object type.

[0131]

[0135] AB. The method of paragraph Z, wherein the radar response threshold includes a first Doppler value associated with the object type and a second radar cross section value associated with the object type.

[0132]

[0136] AC. The method of paragraph Z, wherein determining the radar noise level associated with the second region includes determining a first difference between a first distance from a radar device to the first region and a second distance from the radar device to the second region, determining a second difference in Doppler measurements between the first region and the second region, and determining an intensity measurement associated with the detection.

[0133]

[0137] AD. The method of paragraph Z, wherein determining the radar response threshold is based at least in part on a Doppler distribution associated with the object type.

[0134]

[0138] AE. The method of paragraph Z, wherein determining the radar response thresholds includes determining distance and azimuth values ​​associated with the second region and determining a first radar response threshold and a second radar response threshold based at least in part on the distance and azimuth values, the first radar response threshold being associated with a first Doppler value and the second radar response threshold being associated with a second Doppler value.

[0135]

[0139] AF. The method of paragraph AE, further including determining a first probability associated with the first radar response threshold and determining a second probability associated with the second radar response threshold based at least in part on a Doppler distribution associated with the object type.

[0136]

[0140] AG. The method of paragraph AF, further including determining a first false negative probability associated with the object type based at least in part on the first probability and the first radar response threshold, determining a second false negative probability associated with the object type based at least in part on the second probability and the second radar response threshold, and determining a third false negative probability associated with the second region based at least in part on the first false negative probability and the second false negative probability.

[0137]

[0141] AH. One or more non-transitory computer-readable media storing processor-executable instructions that, when executed by one or more processors, cause the one or more processors to perform operations including receiving radar data indicative of a detection associated with a first region in an environment, the radar data including Doppler data associated with the detection; determining a radar noise level associated with a second region based at least in part on the Doppler data associated with the detection; determining a radar response threshold associated with an object type; and controlling a vehicle in the environment based at least in part on the radar noise level associated with the second region and the radar response threshold associated with the object type.

[0138]

[0142] AI. The one or more non-transitory computer-readable media of paragraph AH, wherein the radar response threshold includes a first Doppler value associated with the object type and a second radar cross section value associated with the object type.

[0139]

[0143] AJ. One or more non-transitory computer-readable media of paragraph AH, wherein determining the radar noise level associated with the second region includes determining a first difference between a first distance from a radar device to the first region and a second distance from the radar device to the second region, determining a second difference in Doppler measurements between the first region and the second region, and determining an intensity measurement associated with the detection.

[0140]

[0144] AK. The one or more non-transitory computer-readable media of paragraph AH, wherein determining the radar response threshold is based at least in part on a Doppler distribution associated with the object type.

[0141]

[0145] AL. The one or more non-transitory computer-readable media of paragraph AH, wherein determining the radar response thresholds includes determining distance values ​​and azimuth values ​​associated with the second region and determining a first radar response threshold and a second radar response threshold based at least in part on the distance values ​​and the azimuth values, wherein the first radar response threshold is associated with a first Doppler value and the second radar response threshold is associated with a second Doppler value.

[0142]

[0146] AM. The one or more non-transitory computer-readable media described in paragraph AL, wherein the operations further include determining a first probability associated with the first radar response threshold and determining a second probability associated with the second radar response threshold based at least in part on a Doppler distribution associated with the object type.

[0143]

[0147] AN. The one or more non-transitory computer-readable media of paragraph A, wherein the operations further include determining a first false negative probability associated with the object type based at least in part on the first probability and the first radar response threshold, determining a second false negative probability associated with the object type based at least in part on the second probability and the second radar response threshold, and determining a third false negative probability associated with the second region based at least in part on the first false negative probability and the second false negative probability.

[0144]

[0148] Although the example sections above are described with respect to particular implementations, it should be understood that in the context of this specification, the contents of the example sections can be implemented via methods, devices, systems, computer-readable media, and / or other implementations. Additionally, any of the examples A through AN may be implemented alone or in combination with any other one or more of A through AN.

[0145] summary

[0149] Although one or more examples of the technology described herein have been described, various modifications, additions, permutations, and equivalents thereof are included within the scope of the technology described herein. As can be understood, the components discussed herein are described as separated for illustrative purposes. However, the operations performed by the various components can be combined or performed in any other components. It should also be understood that components or steps discussed with respect to one example or implementation can be used in combination with components or steps of other examples.

[0146]

[0150] A non-limiting list of objects in the environment may include, but is not limited to, pedestrians, animals, cyclists, trucks, motorcycles, other vehicles, and the like. Such objects in the environment have a "geometric pose" (which may also be referred to herein simply as "pose") that includes the overall object's location and / or orientation relative to a reference frame. In some examples, the pose may indicate the object's (e.g., pedestrian's) location, the object's orientation, or the object's relative appendage positions. The geometric pose may be described in two dimensions (e.g., using an xy coordinate system) or three dimensions (e.g., using an xyz or polar coordinate system) and may include the object's orientation (e.g., roll, pitch, and / or yaw). Some objects, such as pedestrians and animals, also have what is referred to herein as an "appearance pose." Appearance pose includes the shape and / or position of body parts (e.g., appendages, head, torso, eyes, hands, feet, etc.). As used herein, the term "pose" refers to both the "geometric pose" of an object relative to a reference frame, and the "appearance pose" in the case of pedestrians, animals, and other objects that can change the shape and / or position of body parts. In some examples, the reference frame is described with reference to a two-dimensional or three-dimensional coordinate system or map that describes the position of the object relative to the vehicle. However, in other examples, other reference frames may be used.

[0147]

[0151] In the illustrative description, reference is made to the accompanying drawings, which form a part hereof, showing by way of illustration certain examples of the claimed subject matter. It should be understood that other examples can be used and modifications or substitutions, such as structural changes, can be made. Such examples, modifications, or variations do not necessarily depart from the intended scope of the claimed subject matter. Although steps herein may be presented in a certain order, in some cases the order may be changed such that certain inputs are provided at different times or in a different order without changing the functionality and method of the described system. The disclosed procedures may also be performed in different orders. Additionally, the various calculations herein need not be performed in the order disclosed, and other examples using alternative orders of calculations may be readily implemented. In addition to reordering, calculations may also be decomposed into sub-calculations with the same results.

[0148]

[0152] Although the subject matter has been described in language specific to structural features and / or method acts, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described. Rather, the specific features and acts are disclosed as example forms of implementing the claims.

[0149]

[0153] The components described herein represent instructions that may be stored on any type of computer-readable medium and implemented in software and / or hardware. All of the methods and processes described above may be embodied in and fully automated via software code and / or computer-executable instructions executed by one or more computers or processors, hardware, or some combination thereof. Some or all of the methods may alternatively be embodied in specialized computer hardware.

[0150]

[0154] Unless otherwise noted, conditional terms such as "may," "could," "may," or "might," among others, are understood within the context to indicate that an example includes certain features, elements, and / or steps that other examples do not include. Thus, such conditional terms are not intended to generally imply that a certain feature, element, and / or step is required in any way in one or more examples, or that the one or more examples necessarily include logic for determining whether a certain feature, element, and / or step is included in or should be performed in any particular example, with or without user input or prompting.

[0151]

[0155] Connecting language such as the phrase "at least one of X, Y, or Z" should be understood to mean that the item, term, etc. can be either X, Y, or Z, or any combination thereof, including collections of each of the elements, unless otherwise noted. Unless expressly stated as singular, "a" means singular and plural.

[0152]

[0156] The descriptions, elements or blocks of the routines in the flow diagrams described herein and / or depicted in the accompanying drawings should be understood to potentially represent modules, segments or portions of code that include one or more computer-executable instructions for implementing a particular logical function or element within the routine. Included within the scope of the examples described herein are alternative implementations that remove elements or functions, or perform operations substantially synchronously, in reverse order, with additional operations, or in an order different from that shown or described, such as omitting operations, depending on the functionality involved as would be understood by one of ordinary skill in the art.

[0153]

[0157] Many variations and modifications may be made to the examples described above, and elements thereof are understood to be included in other acceptable examples. All such modifications and variations are intended to be included herein within the scope of this disclosure and protected by the following claims.

Claims

1. A system comprising one or more processors and one or more non - transient computer - readable media storing computer - executable instructions, wherein when the instructions are executed, the system is caused to receive radar data associated with a region in an environment; determine radar noise data associated with the region; evaluate the radar data based at least in part on a radar response threshold of a vehicle and the radar noise data, wherein the radar response threshold is based at least in part on a vehicle safety metric; control a vehicle in the environment based at least in part on the radar response threshold and the radar data; perform operations including the above.

2. The vehicle safety metric is associated with a first object type, and the operations further include determining a probability associated with detecting an object of the first object type at a first position in the environment based at least in part on the radar noise data and a radar response distribution associated with the first object type; determining the radar response threshold based at least in part on the probability; The system according to claim 1.

3. The operations further include determining a second radar response threshold associated with a second object type based at least in part on the radar noise data and a second radar response distribution associated with the second object type; determining a second probability associated with detecting a second object of the second object type at the first position based at least in part on the radar noise data and the second radar response distribution; The system according to claim 2.

4. The operations further include determining a drivable surface metric associated with operating the vehicle in the environment; determining the radar response threshold based at least in part on the drivable surface metric; The system according to claim 1 or 2.

5. The operations include Determining a first radar noise level associated with a first position in the environment and a second radar noise level associated with a second position in the environment, at least partially based on the radar noise data; Excluding the first position and including the second position, a trajectory of the vehicle; Determining that the first radar noise level meets or exceeds the radar response threshold; and Determining that the second radar noise level is below the radar response threshold; Generating, at least partially based on; The system according to claim 1 or 2, further comprising.

6. Determining the radar noise data includes: Determining a first position in the environment associated with the radar data; and Determining a radar noise level associated with a second position in the environment, at least partially based on a difference in distance from a radar device between the first position and the second position and an intensity measurement associated with the radar data; The system according to claim 1 or 2, comprising.

7. Receiving radar data associated with an area in the environment; Determining radar noise data associated with the area; Evaluating the radar data, at least partially based on a radar response threshold of the vehicle and the radar noise data, wherein the radar response threshold is at least partially based on a vehicle safety metric; Controlling a vehicle in the environment, at least partially based on the radar response threshold and the radar data; A method comprising.

8. The vehicle safety metric is associated with a first object type, and the method includes: Determining a probability associated with detecting an object of the first object type at a first position in the environment, at least partially based on the radar noise data and a radar response distribution associated with the first object type; Determining the radar response threshold, at least partially based on the probability; The method according to claim 7, further comprising.

9. Determining a second radar response threshold associated with a second object type based at least in part on the radar noise data and a second radar response distribution associated with the second object type; Determining a second probability associated with detecting a second object of the second object type at the first position based at least in part on the radar noise data and the second radar response distribution; The method according to claim 8, further comprising. **Claim 10** Determining a first radar response threshold; Determining a first number of false positive radar detections and a first number of false negative radar detections associated with the first radar response threshold; Determining a second radar response threshold; Determining a second number of false positive radar detections and a second number of false negative radar detections associated with the second radar response threshold; Determining either the first radar response threshold or the second radar response threshold as the radar response threshold based at least in part on the first number of false positive radar detections and the second number of false positive radar detections, and the first number of false negative radar detections and the second number of false negative radar detections; The method according to claim 7 or claim 8, further comprising. **Claim 11** Determining a drivable surface metric associated with operating the vehicle in the environment; Determining the radar response threshold based at least in part on the drivable surface metric; The method according to claim 7 or 8, further comprising. **Claim 12** Determining a first radar noise level associated with a first position in the environment and a second radar noise level associated with a second position in the environment based at least in part on the radar noise data; Excluding the first position and including the second position, a trajectory of the vehicle; Determining that the first radar noise level meets or exceeds the radar response threshold; and Determining that the second radar noise level is below the radar response threshold; Generating based at least in part on; The method according to claim 7 or 8, further comprising. **Claim 13** Determining the radar noise data comprises Determining the first position in the environment associated with the radar data; Determining a radar noise level associated with a second position in the environment, based at least in part on a difference in distance from a radar device between the first position and the second position, and on an intensity measurement associated with the radar data; The method according to claim 7 or 8, comprising.

14. Determining a first vehicle safety metric associated with a first environment; Determining a second vehicle safety metric associated with a second environment; Determining the operating environment of the vehicle; Determining, based at least in part on the operating environment of the vehicle, either the first vehicle safety metric or the second vehicle safety metric as the vehicle safety metric; The method according to claim 7 or 8, further comprising.

15. One or more non-transitory computer-readable media comprising instructions that, when executed by one or more processors, cause the one or more processors to perform the method according to claim 7 or 8; Non-transitory computer-readable media.

Images