Methods for navigating vehicles and systems for navigating vehicles

Abstract

The vehicle includes a system and method for navigating the vehicle. The system includes sensors and a processor. The sensors capture images of the road. The processor uses a machine learning program to focus the sensors on road segments selected from multiple road segments based on the risk of each segment. The machine learning program is trained to focus the sensors by: calculating the risk of each of the multiple road segments based on the hazard probability and occupancy probability associated with each road segment; selecting road segments from the multiple road segments based on the risks associated with each segment; and determining the reduction in risk in the road risk model resulting from the selection of said road segments.

Description

Technical Field

[0001] This subject matter discloses methods for mitigating the risks of vehicles entering road segments, and in particular for systems and methods for calculating the risks of intersections and operating vehicles based on the calculated risks. Background Technology

[0002] When vehicles pass through intersections, the risk of accidents increases due to factors such as intersecting traffic, changes in speed, changes in lighting conditions, and pedestrians crossing at intersections. Therefore, autonomous vehicles entering intersections require additional measurements of the surrounding traffic conditions to ensure successful passage. For effectiveness, these measurements and subsequent calculations need to be performed in real time. Therefore, a simple method is desired to assess the risk of a vehicle passing through an intersection based on current traffic conditions. Summary of the Invention

[0003] In one exemplary embodiment, a method for navigating a vehicle is disclosed. Images of a road are acquired from sensors. The sensors focus on road segments selected from a plurality of road segments, wherein the road segments are selected based on the risk of each road segment using a machine learning program. The machine learning program is trained to select road segments by calculating the risk of each of the plurality of road segments, wherein the risk associated with a road segment is based on the probability of danger associated with the road segment and the probability of occupancy associated with the road segment, selecting road segments from the plurality of road segments based on the risk associated with the road segments, and determining the risk reduction of the road's road risk model due to the selection of road segments.

[0004] In addition to one or more features described herein, calculating the risk of a road segment also includes calculating the product of the hazard probability of the road segment and the occupancy probability of the road segment. The focusing sensor also includes performing a random selection process over multiple road segments, wherein the probability of selecting a road segment is based on the risk associated with that road segment. The method also includes warning the driver when the driver's attention is not on the road segment. The method also includes using tactile signals to direct the driver's attention to the road segment. The method also includes comparing the risk of the road risk model with a risk metric, and rewarding a machine learning program for selecting the road segment when the risk is less than the risk metric. In one embodiment, the sensor includes a first sensor and a second sensor, and the method also includes focusing the first sensor on the road segment while maintaining a wide field of view of the road with the second sensor.

[0005] In another exemplary embodiment, a system for navigating a vehicle is disclosed. The system includes a sensor and a processor. The sensor is configured to capture images of a road. The processor is configured to focus the sensor on a road segment selected from a plurality of road segments using a machine learning program based on the risk of the road segment. The machine learning program is trained to focus the sensor by calculating the risk of each of the plurality of road segments, wherein the risk associated with a road segment is based on a hazard probability associated with the road segment and an occupancy probability associated with the road segment, the road segment is selected from the plurality of road segments based on the risk associated with the road segment, and a reduction in the risk of the road's road risk model resulting from the selection of said road segment is determined.

[0006] In addition to one or more features described herein, the processor is configured to calculate the risk of a road segment by multiplying the hazard probability of the road segment by the occupancy probability of the road segment. The processor is also configured to focus the sensor on a plurality of road segments by performing a random selection process, wherein the probability of selecting a road segment is based on the risk associated with that road segment. The processor is further configured to warn a driver when the driver's attention is not on the road segment. The processor is also configured to use tactile signals to direct the driver's attention to the road segment. The processor is further configured to train a machine learning program by comparing the risk of a road risk model with a risk metric, and to reward the machine learning program for selecting the road segment when the risk is less than the risk metric. In one embodiment, the sensor includes a first sensor and a second sensor, and the processor is further configured to focus the first sensor on the road segment while maintaining a wide field of view of the road with the second sensor.

[0007] In yet another exemplary embodiment, a vehicle is disclosed. The vehicle includes a sensor and a processor. The sensor is configured to capture images of a road. The processor is configured to focus the sensor on a road segment selected from a plurality of road segments using a machine learning program based on the risk of the road segment. The machine learning program is trained to focus the sensor by calculating the risk of each of the plurality of road segments, wherein the risk associated with a road segment is based on a hazard probability associated with the road segment and an occupancy probability associated with the road segment, the road segment is selected from the plurality of road segments based on the risk associated with the road segment, and a reduction in the risk of the road's road risk model resulting from the selection of said road segment is determined.

[0008] In addition to one or more features described herein, the processor is configured to calculate the risk of a road segment by multiplying the hazard probability of the road segment by the occupancy probability of the road segment. The processor is also configured to focus the sensor on a plurality of road segments by performing a random selection process, wherein the probability of selecting a road segment is based on the risk associated with that road segment. The processor is further configured to use tactile signals to redirect the driver's attention to the road segment when the driver's attention is not on the road segment. The processor is also configured to train a machine learning program by comparing the risk of a road risk model with a risk metric, and to reward the machine learning program for selecting the road segment when the risk is less than the risk metric. In one embodiment, the sensor includes a first sensor and a second sensor, and the processor is further configured to focus the first sensor on the road segment while maintaining a wide field of view of the road with the second sensor.

[0009] The above-described features and advantages, as well as other features and advantages, of this disclosure will become apparent from the following detailed description when taken in conjunction with the accompanying drawings. Attached Figure Description

[0010] Other features, advantages, and details appear only by way of example in the following detailed description, which refers to the accompanying drawings, wherein:

[0011] Figure 1 A vehicle according to an exemplary embodiment is shown;

[0012] Figure 2 An example of using Figure 1 The vehicle's driver alert system;

[0013] Figure 3 A top view of an intersection in an illustrative embodiment is shown;

[0014] Figure 4 It shows when the main vehicle approaches Figure 3 A complete probability diagram of danger for the main vehicle at an intersection;

[0015] Figure 5 It shows when the main vehicle is at the intersection Figure 3 Danger probability diagram of an intersection;

[0016] Figure 6 It shows Figure 3 The occupancy probability map of the intersection;

[0017] Figure 7 The image shows a schematic view of an intersection as a primary vehicle approaches the intersection;

[0018] Figure 8 A close-up of the selected road segment at the intersection is shown;

[0019] Figure 9 A block diagram is shown in one embodiment of the process for training a machine learning program to focus a sensor on a road segment;

[0020] Figure 10 A flowchart is shown for a method of alerting passengers or directing vehicles to the risks within an intersection; and

[0021] Figure 11 A flowchart is shown for a method for determining the risk of an intersection or road segment, including the possibility that an object may be obscured from the field of view of a sensing system. Detailed Implementation

[0022] The following description is exemplary in nature only and is not intended to limit this disclosure, its application, or use. It should be understood that in all the drawings, corresponding reference numerals denote the same or corresponding parts and features.

[0023] According to an exemplary embodiment, Figure 1 Vehicle 10 is illustrated. In various embodiments, vehicle 10 may be autonomous or semi-autonomous. In an exemplary embodiment, vehicle 10 is a so-called Level 4 or Level 5 automation system. A Level 4 system signifies “high automation,” referring to the autonomous driving system’s driving modes—specific performance—in all aspects of a dynamic driving task, even if the human driver does not respond appropriately to intervention requests. A Level 5 system signifies “full automation,” referring to the autonomous driving system’s full-time performance in all aspects of a dynamic driving task under all road and environmental conditions that a human driver can manage. It should be understood that the systems and methods disclosed herein can also be used with vehicles operating at any level from Level 1 to Level 5.

[0024] Vehicle 10 typically includes at least a navigation system 20, a propulsion system 22, a transmission system 24, a steering system 26, a braking system 28, a sensing system 30, an actuator system 32, and a controller 34. The navigation system 20 determines road-level route plans for the autonomous driving of vehicle 10. The propulsion system 22 provides power to generate prime mover for vehicle 10 and, in various embodiments, may include an internal combustion engine, an electric motor such as a traction motor, and / or a fuel cell propulsion system. The transmission system 24 is configured to transmit power from the propulsion system 22 to two or more wheels 16 of vehicle 10 according to a selectable speed ratio. The steering system 26 affects the position of two or more wheels 16. Although depicted as including a steering wheel 27 for illustrative purposes, in some embodiments contemplated within the scope of this disclosure, the steering system 26 may not include a steering wheel 27. The braking system 28 is configured to provide braking torque to two or more wheels 16.

[0025] The sensing system 30 includes sensors or detectors that sense objects 50 in the external environment of the vehicle 10 and determine various parameters of the objects, which are useful for locating the position and relative speed of various remote vehicles in the autonomous vehicle environment. These parameters may be provided to the controller 34. The sensing system 30 may include a first sensor and a second sensor. In various embodiments, the first sensor may be used to focus on a selected segment of a road or intersection, while the second sensor is used to obtain a wide field of view of the road or intersection. The first sensor may narrow its field of view and may also change its orientation. In one embodiment, the sensing system 30 includes one or more digital cameras for capturing one or more images of the road or intersection. In alternative embodiments, the sensing system 30 may include one or more radar systems, lidar, etc., for detecting the distance, relative speed, azimuth, and elevation angle of objects 50 such as target vehicles, pedestrians, etc.

[0026] The controller 34 includes a processor 36 and a computer-readable storage device or computer-readable storage medium 38. The storage medium includes a program or instructions 39 that, when executed by the processor 36, operate the vehicle 10 based on outputs from the sensing system 30. The controller 34 can construct a trajectory of the vehicle 10 based on the outputs of the sensing system 30 and can provide the trajectory to the actuator system 32 to control the propulsion system 22, the transmission system 24, the steering system 26, and / or the braking system 28, thereby navigating the vehicle 10 relative to the object 50.

[0027] The computer-readable storage medium 38 may also include a program or instructions 39 that, when executed by the processor 36, determine the risk to vehicle 10 in a traffic scenario, particularly at an intersection. The controller 34 may provide alerts to the driver of vehicle 10 and / or control the vehicle's operation based on the risk. In various embodiments, the controller 34 may also accelerate and / or decelerate vehicle 10, steer the vehicle, apply brakes, etc., to avoid a collision or impact with object 50.

[0028] The computer-readable storage medium 38 may further include a program or instructions 39 that, when executed by the processor 36, focus a sensor on a selected segment of the road based on the risk associated with that segment. A machine learning program (e.g., a neural network) can be used to focus the sensor. The machine learning program can be trained to focus the sensor using real or simulated data, as described herein.

[0029] Figure 2 An example of using Figure 1The vehicle 10 includes a driver alert system 200. The driver alert system 200 includes a driver monitoring device 202 that observes the driver and the driver's level of attention. In various embodiments, the driver monitoring device 202 may be an eye monitor or a biosensor indicating the driver's level or direction of attention. Data from the driver monitoring device 202 is provided to a controller 34. The controller 34 communicates with the steering system 26 and / or the driver's seat 204. The controller 34 compares the focus of the sensing system (e.g., the focus of one of the first and second sensors) and provides an alert to the driver when the driver's attention is not aligned with the focus of the sensing system 30. The controller 34 may issue an alert or provide a tactile warning, such as by vibrating an object in contact with the driver, such as the steering wheel 27 or the driver's seat 204. The driver's seat 204 may have tactile emitters at different locations along the driver's seat 204. When the driver's attention is not aligned with the focus of the sensing system 30, a tactile emitter corresponding to the focus direction of the sensing system 30 can be selected. For example, the right-side tactile emitter can be activated to direct the driver's attention to the right, and the left-side tactile emitter can be activated to direct the driver's attention to the left.

[0030] Figure 3 A top view 300 of an intersection 302 in an illustrative embodiment is shown. In various embodiments, the intersection 302 can have any configuration of suitable traffic indicators, such as four-way stop signs, two-way stop signs, or traffic lights. For illustrative purposes, the top view 300 includes the main vehicle 301 (e.g., Figure 1 Vehicle 10) passes through its selected lane 304 approaching intersection 302, the oncoming lane 306 which allows traffic flow to move in the opposite direction to the selected lane, the first intersecting traffic lane 308, and the second intersecting traffic lane 310. Given a stop sign or red light condition in the selected lane 304, the stop line 312 in the selected lane 304 indicates where the main vehicle 301 should stop when approaching intersection 302.

[0031] Figure 3 The model used to determine the hazard probability within the selected lane 304 is also partially illustrated. The controller 34 divides the selected lane 304 into multiple segments according to the segmentation algorithm model and assigns a hazard probability to each segment. The hazard probability indicates the probability that the main vehicle 301 will encounter an accident or hazard when an object 50, such as the target vehicle, is in the segment. The hazard probability does not require the object 50 to be present in the segment, and the value of the hazard probability is independent of whether the object is in the segment. The selected lane 304 shows multiple segments ahead of the vehicle 10.

[0032] Each road segment is shaded to indicate the range of hazard probabilities associated with that segment. For example, the first guide segment 314, immediately in front of the main vehicle 301, has a high hazard probability associated with it because there is a high probability of collision if a target vehicle is in it. The second guide segment 316 is further away from the main vehicle 301 and has a medium hazard probability associated with it, mainly because the main vehicle 301 has extra time to react to a target vehicle in it. Similarly, the third guide segment 318 has a low hazard probability associated with it, and the fourth guide segment 320 has a very low hazard probability associated with it. In various embodiments, the high hazard probability is between approximately 0.75 and 1, the medium hazard probability is between approximately 0.5 and approximately 0.75, the low hazard probability is between approximately 0.25 and 0.5, and the very low hazard probability is between 0 and approximately 0.25.

[0033] Figure 4 This shows that when the main vehicle approaches intersection 302, the main vehicle 301 targets... Figure 3 The complete hazard probability map 400 for intersection 302. Each of the selected lanes 304, oncoming lanes 306, and first intersecting lanes 308 and second intersecting lanes 310 is divided by controller 34 and assigned an associated hazard probability.

[0034] like Figure 4 As shown, a segment in a lane (e.g., the second guide segment 416 in selected lane 304) may overlap with a segment in an intersecting lane (e.g., the third intersecting segment 406 in the first intersecting traffic lane 308). It should be understood that each of these overlapping segments may have a different probability of danger due to the relative direction of traffic associated with each segment.

[0035] Similarly, Figure 4 As shown, the hazard probability of a road segment can be based on the speed of the target vehicle within that segment and the current state of the main vehicle 301. The state of the main vehicle 301 includes its position relative to the intersection and its current speed. When determining the hazard probability, the hazard probability model can also consider various possible maneuvers of the main vehicle 301, such as turning left, turning right, or proceeding straight through the intersection. The hazard probability can be pre-calculated before the main vehicle 301 reaches the intersection 302.

[0036] The first intersection 402, the second intersection 404, the third intersection 406, and the fourth intersection 408 illustrate the effect of the assumed target vehicle speed on the probability of danger. Regarding the first intersection 402, although this section is far from the intersection, the target vehicle in this section, if at the assumed speed, would be on the collision path with the main vehicle 301.

[0037] The second intersection section 404 has a medium risk probability because the likelihood of a target vehicle with a presumed speed colliding with the host vehicle 301 in this section is relatively low. Similarly, the third intersection section 406 has a low risk probability because when the host vehicle reaches this section, the target vehicle in this section is most likely to move out of the path of the host vehicle 301. Similarly, the fourth intersection section 408 has a very low risk probability because the host vehicle 301 is relatively inaccessible.

[0038] Figure 5 Fig. 500 shows the risk probability diagram of the intersection ۳۰۲ when the host vehicle ۳۰۱ is at the intersection ۳۰۲. Figure 3 of the intersection 302. Figure 4 and Figure 5 The comparison shows how the risk probability of each section changes based on the state of the host vehicle 301. Although the risk probability of the first intersection section 402 remains high, the probability of the second intersection section 404 has changed from medium ( Figure 4 in Figure 5 [[ID=1〗) to high ( Figure 4 in Figure 5 [[ID=18〗). Similarly, the risk probability of the third intersection section 406 has changed from low ( Figure 4 in Figure 5 [[ID=22〗) to medium (

[0039] Figure 6 shows Figure 3 the occupancy probability diagram 600 of the intersection 302. The occupancy probability diagram 600 is based on the detections made by the sensing system 30. Figure 6 The occupancy probability diagram 600 shown in

[0040] Once the risk probability and the occupancy probability are known, the risk for each section and the entire intersection can be calculated. The risk of the selected section (the nth section) is the product of the occupancy probability of that section and the risk probability of that section, as shown in Equation (1):

[0041] Risk

[0041] , n , n ,

[0042] , , n , n , = P(O [[ID=3〗)×P(C n ) Equation (1)

[0042] where P(O n ​P(C) is the occupancy probability of the nth road segment. n ) is the danger probability of the nth road segment. The risk of the entire intersection is the sum of the risks of the multiple road segments of the intersection, as shown in equation (2).

[0043]

[0044] Where N is the total number of road segments. The identified risk of an intersection can be compared with a warning threshold to signal the processor or driver, thereby preventing the main vehicle 301 from entering the intersection where an accident may occur.

[0045] Figure 7 Image 700 illustrates an illustrative intersection as seen by the primary vehicle 301 approaching the intersection. Image 700 shows a first target vehicle 702 in the oncoming lane and a second target vehicle 704 in the intersecting lane. Image 700 also includes a traffic light 706. Controller 34 isolates the first target vehicle 702 and the second target vehicle 704 using a bounding box, which can then be used to identify the vehicles. Controller 34 also isolates the traffic light 706 to aid in identifying the traffic light's status.

[0046] Figure 8 A close-up of a selected road segment at the intersection is shown. In various embodiments, controller 34 identifies the road segment with the highest relevance risk (e.g., the segment for the second target vehicle 704) and then focuses the field of view of sensing system 30 on the high-risk road segment to mitigate the risk. Focusing can be performed using one of the first and second sensors of sensing system 30. While one of the first and second sensors is focused on the selected road segment, the other maintains a wide field of view of the intersection. By focusing the sensor on the selected road segment, controller 34 can obtain an updated occupancy probability and risk value for the selected road segment. The updated occupancy probability can be used to update the risk of the entire intersection.

[0047] Figure 9 A block diagram 900 is shown in one embodiment of a process for training a machine learning program to focus sensors on road segments. The process acquires input from various sensors and determines road features based on the input. These features are fed into a machine learning program that executes a random selection strategy to select road segments for sensor focus. A road risk model is then determined based on the sensor focus. The road risk model defines the risks associated with the entire road and can be used to train the machine learning program.

[0048] The inputs used to train the machine learning program can come from many sources. The road importance model 902 and the intersection map 904 can be provided from a database or a remote server. Camera data 906 and other sensor data 908 can come from various components of the sensing system 30, such as digital cameras 40 and / or radar, lidar, etc. For example, the salience of the intersection 910 and previous focal position data 912 can be stored in a computer-readable storage medium 38 or a controller 34.

[0049] A road importance model 902 and an intersection map 904 are used to determine a wide-field image 914 of the road intersection, which has weights from a hazard probability map. Camera data 906 and other sensor data 908 are used to perform road segmentation 916 and determine previous vehicle detections 918 of objects in the road, as well as their positions and speeds. The saliency of the intersection 910 and previous focal point location data 912 are used to create an uncertainty map 920, which can be used to train a machine learning program 922.

[0050] The machine learning program 922 can be a neural network, a Gaussian process machine, a support vector machine, or other suitable machine learning program. A wide-field-of-view image 914, road segments 916, previous vehicle detections 918, and an uncertainty map 920 are provided to the machine learning program 922. The machine learning program 922 receives these features and training data 948 and implements a stochastic strategy 924 to select road segments of interest to the sensors. The stochastic strategy 924 includes randomly selecting road segments based on their weights. The risk of a road segment is used as the weight in the random selection process. The probability of selecting a road segment using the random selection process is related to the risk associated with that road segment. For example, a road segment with a high associated risk is more likely to be selected than a road segment with a low associated risk.

[0051] The random selection process outputs a selected road segment 926 for closer focusing. A simulation 930 is then performed to determine a road risk model based on the selection of the chosen road segment. In the simulation, the first sensor focuses on the selected road segment to obtain a narrow FOV camera image 932. The second sensor remains on a wide field of view of the simulated intersection to obtain a wide FOV camera image 934. Generally, a narrow FOV image is a high-resolution image, while a wide FOV image is a low-resolution image.

[0052] A first sensor observes a narrow FOV to obtain vehicle detection 936 for any narrow FOV. A second sensor observes a wide FOV to obtain vehicle detection 938 for any wide FOV. The narrow FOV vehicle detection 936 and the wide FOV vehicle detection 938 are used at an occupancy grid calculator 940 to calculate a focused occupancy grid 942 and a non-focused occupancy grid 944 for the intersection. The values ​​of the focused occupancy grid 942 and the non-focused occupancy grid 944 are used to update the road risk model 946 of the intersection, which can be used as training data 948 in subsequent iterations of the training steps of the machine learning program 922. The road risk model 946 indicates whether focusing the sensor on a selected road segment via the focused occupancy grid 942 reduces or diminishes the risk relative to the non-focused occupancy grid 944. The machine learning program 922 can be rewarded when focusing the sensor on the selected road segment reduces the risk associated with the road risk model, or penalized when focusing the sensor increases the risk. In various embodiments, the risk associated with the road risk model can be compared to a risk metric, and the machine learning program can be rewarded when the risk due to focusing the sensor is less than the risk metric.

[0053] Figure 10 A flowchart 1000 illustrates a method for warning passengers or the driver of a main vehicle 301 of risks within an intersection. In box 1002, the occupancy probability of a road segment is determined. In box 1004, the collision probability of a road segment is determined. The occupancy probability and collision probability are provided to a focusing model 1006 to determine the risk associated with the road segment. The focusing model 1006 identifies selected road segments of interest with associated high risk. Sensors are typically focused on the selected road segments. A machine learning program 922, which has been trained, operates the focusing model 1006 to select the road segments for sensor focusing.

[0054] In box 1008, driver monitoring is performed to determine where the driver's attention is focused. Driver monitoring can be performed using, for example, eye sensors that track the position or direction of the driver's eyes.

[0055] In box 1010, the selected road segment and driver attention are provided to a difference mapping module, which determines the difference between the selected road segment and the driver's focus or attention. When a difference exists between the driver's focus and the selected road segment (i.e., when the driver does not notice the road segment with the greatest risk), an awareness signal can be generated to warn the driver.

[0056] In block 1012, the selected road segment is mapped to a haptic actuator. A haptic signal can be assigned to one or more haptic actuators, for example, on the driver's seat. Selected haptic signals can be sent to focus the driver's attention on a selected location within the road. For example, the haptic actuators may include a first vibrating device located on the left side of the driver's seat and a second device located on the right side of the driver's seat. The first vibrating device can be actuated when the selected road segment is on the driver's left, and the second vibrating device can be actuated when the selected road segment is on the driver's right. Furthermore, the intensity of the haptic signal can be high to indicate a high level of risk and problem, and low to indicate a low level of risk and problem. In block 1014, the mapped haptic signal is sent to a haptic controller to actuate the corresponding haptic actuator.

[0057] Figure 11 A flowchart 1100 illustrates a method for determining the risk of an intersection or road segment, including the probability that an object will be occluded from the field of view of a sensing system 30. In box 1102, a prior belief about the intersection state is generated. In box 1104, observations from the sensors are provided. Based on the prior belief and the observations, the probability of occlusion is determined. In box 1106, the occlusion probability is updated using a Bayesian update process based on the prior belief and the observations.

[0058] In box 1108, for unobstructed scenarios, unobstructed probabilities are sent along the processing path, while for occluded scenarios, occluded probabilities are sent along the processing path. These processing paths run in parallel with each other. Along the processing path for unobstructed scenarios, in box 1110, a system dynamics model is run on the intersection under the assumption of no occlusion. In box 1112, based on the system dynamics model, an updated belief in the intersection state is obtained. In box 1114, the updated state is multiplied by the unobstructed probability to generate a weighted unobstructed state.

[0059] Similarly, following the processing path for the occlusion scenario, in box 1116, a system dynamics model is run on the intersection assuming occlusion. In box 1118, an updated belief in the intersection state is obtained based on the system dynamics model. In box 1120, the updated state is multiplied by the occlusion probability to generate a weighted occlusion state. In summation box 1122, the weighted unoccluded state and the weighted occluded state are summed to determine the risk associated with the selected road segment in box 1124.

[0060] While this method has been discussed for vehicles approaching intersections, it can also be applied to other types of roads, such as straight sections, branching roads, unbranched roads, and curves. Risks can be configured based on any type of intersection, and calculations can include the impact of obstacles (such as buildings, hedges, or hills) on the occurrence of sensor detections.

[0061] While the foregoing disclosure has been described with reference to exemplary embodiments, those skilled in the art will understand that various changes can be made and equivalents can replace its elements without departing from its scope. Furthermore, many modifications can be made to adapt particular situations or materials to the teachings of this disclosure without departing from its essential scope. Therefore, it is intended that this disclosure be limited to the specific embodiments disclosed, but will include all embodiments falling within its scope.

Claims

1. A method for navigating a vehicle, comprising: Images of the road at the intersection are obtained from sensors, including a first sensor and a second sensor; and Determine the hazard probability map of the intersection; Determine the occupancy probability map of the intersection; The first sensor of the sensors is focused on a road segment selected from multiple road segments of the road, wherein the first sensor observes a narrow FOV to obtain vehicle detection for any narrow FOV, and the second sensor observes a wide FOV to obtain vehicle detection for any wide FOV, wherein the road segment is selected based on the risk of the road segment using a machine learning program, which is trained to select road segments in the following manner: Calculate the risk of each of the multiple road segments, where the risk associated with a segment is based on the probability of danger associated with that segment and the probability of occupancy associated with that segment; Based on the risks associated with a road segment, a road segment is selected from multiple road segments; and Determine the risk reduction in the road risk model resulting from the selection of the aforementioned road segment; Specifically, narrow FOV vehicle detection and wide FOV vehicle detection are used at the occupancy grid calculator to calculate the focused occupancy grid and the non-focused occupancy grid of the intersection. The values ​​of the focused occupancy grid and the non-focused occupancy grid are used to update the road risk model of the intersection.

2. The method of claim 1, wherein focusing the sensor further comprises performing a random selection process on the plurality of road segments, wherein the probability of selecting a road segment is based on the risk associated with the road segment.

3. The method of claim 1 further includes warning the driver when the driver's attention is not on the road.

4. The method of claim 1, further comprising comparing the risk of the road risk model with a risk metric, and rewarding the machine learning program for selecting the road segment when the risk is less than the risk metric.

5. A system for navigating a vehicle, comprising: A sensor, configured to capture images of roads at an intersection, includes a first sensor and a second sensor; and The processor is configured as follows: Determine the hazard probability map of the intersection; Determine the occupancy probability map of the intersection; Based on the risk of road segments, a machine learning program is used to focus sensors on selected road segments from multiple road segments. The machine learning program is trained to focus the sensors in the following way: Calculate the risk of each of the multiple road segments, where the risk associated with that road segment is based on the probability of danger associated with that road segment and the probability of occupancy associated with that road segment; Select a road segment from multiple road segments based on the risks associated with that road segment; and Determine the risk reduction in the road risk model due to the selection of road segments; The first sensor observes a narrow FOV to obtain vehicle detection for any narrow FOV, and the second sensor observes a wide FOV to obtain vehicle detection for any wide FOV. The narrow FOV vehicle detection and the wide FOV vehicle detection are used at the occupancy grid calculator to calculate the focused occupancy grid and the non-focused occupancy grid of the intersection. The values ​​of the focused occupancy grid and the non-focused occupancy grid are used to update the road risk model of the intersection.

6. The system according to claim 5, wherein, The processor is also configured to focus the sensor by performing a random selection process on the plurality of road segments, wherein the probability of selecting a road segment is based on the risk associated with the road segment.

7. The system according to claim 5, wherein, The processor is also configured to alert the vehicle driver when the driver's attention is not on the road segment.

8. The system according to claim 5, wherein, The processor is also configured to train the machine learning program by comparing the risk of the road risk model with a risk metric, and to reward the machine learning program for selecting the road segment when the risk is less than the risk metric.

Images