Method of operating a vehicle and system for operating a vehicle

Abstract

The present invention discloses a spatially invariant three-dimensional convolutional network under spherical coordinate input. A vehicle and a system and method for operating the vehicle, the system comprising a sensor and a processor. The sensor obtains a detection set associated with an object at a certain range from the vehicle. The processor is configured to select a convolution path for the detection set according to the range of the object, wherein the convolution path includes one or more convolution layers, and the number of the one or more convolution layers depends on the range of the object; apply the one or more convolution layers of the selected convolution path to the detection set to generate a filtered data set; and operate the vehicle relative to the object using the filtered data set.

Description

Technical Field

[0001] The present disclosure relates to a system and method for determining characteristics of an object relative to a vehicle, and more particularly to a system and method for adjusting a convolution filter based on the distance of the object from the vehicle. Background Art

[0002] An autonomous vehicle detects objects in its environment and navigates along a selected trajectory relative to the object. One or more sensors associated with the autonomous vehicle may be used to detect objects. Three-dimensional sensors such as lidar and radar essentially provide point cloud detections that can be provided to a convolutional network to extract features of the object. Since the point cloud detections are received in a spherical coordinate system, the number of detections received from an object may vary significantly with distance. Applying a standard convolutional network to such detections is problematic because convolutional networks rely on spatially invariant features. In addition, converting the detections to Cartesian coordinates may produce a point density that varies with range. Therefore, it is desirable to provide a convolutional network that can provide an output that does not vary spatially with range. Summary of the invention

[0003] In one exemplary embodiment, a method of operating a vehicle is disclosed. A detection set associated with an object at a range from the vehicle is obtained. A convolutional path of the detection set is selected based on the range of the object, wherein the convolutional path includes one or more convolutional layers, and the number of the one or more convolutional layers depends on the range of the object. The one or more convolutional layers of the selected convolutional path are applied to the detection set to generate a filtered data set. The vehicle is operated relative to the object using the filtered data set.

[0004] In addition to one or more features described herein, the one or more convolutional layers include a first convolutional layer, and the sampling rate of the first convolutional layer increases as the range decreases. In one embodiment where the one or more convolutional layers include the first convolutional layer, the method further includes defining a first subrange relative to the vehicle and a second subrange relative to the vehicle, wherein the second subrange is smaller than the first subrange; applying the first convolutional layer at a first sampling rate when the object is within the first subrange; and applying the first convolutional layer at a second sampling rate when the object is within the second subrange, wherein the second sampling rate is twice the first sampling rate. The method further includes applying the second convolutional layer to the intermediate set generated in the first convolutional layer at the first sampling rate when the object is within the second subrange. The method further includes defining a third subrange relative to the vehicle, wherein the third subrange is smaller than the second subrange; applying the first convolutional layer at a third sampling rate when the object is within the third subrange, wherein the third sampling rate is twice the second sampling rate; applying the second convolutional layer to the intermediate set generated in the first convolutional layer at the second sampling rate; and applying the third convolutional layer to the output of the second convolutional layer at the first sampling rate. In another embodiment where the one or more convolutional layers include a first convolutional layer, the method further includes defining a first subrange relative to the vehicle and a second subrange relative to the vehicle, wherein the second subrange is smaller than the first subrange; and one of applying the first convolutional layer to the detection set once within the first subrange and applying the first convolutional layer twice within the second subrange, wherein the sampling rate of the first convolutional layer is the same within the first subrange and within the second subrange. The method further includes defining the first subrange and the second subrange, and defining a first weight for the first subrange and a second weight for the second subrange to select a convolution path, wherein at a radial boundary between the first subrange and the second subrange, the first weight and the second weight decrease linearly with radial distance, wherein the sum of the first weight and the second weight is equal to 1.

[0005] In another exemplary embodiment, a system for operating a vehicle is disclosed. The system includes a sensor and a processor. The sensor obtains a detection set associated with an object at a range from the vehicle. The processor is configured to select a convolutional path for the detection set based on the range of the object, wherein the convolutional path includes one or more convolutional layers, and the number of the one or more convolutional layers depends on the range of the object; apply the one or more convolutional layers of the selected convolutional path to the detection set to generate a filtered data set; and operate the vehicle relative to the object using the filtered data set.

[0006] In addition to one or more features described herein, the one or more convolutional layers include a first convolutional layer, and the sampling rate of the first convolutional layer increases as the range decreases. In one embodiment where the one or more convolutional layers include the first convolutional layer, the processor is further configured to define a first sub-range relative to the vehicle and a second sub-range relative to the vehicle, wherein the second sub-range is smaller than the first sub-range; apply the first convolutional layer at a first sampling rate when the object is within the first sub-range; and apply the first convolutional layer at a second sampling rate when the object is within the second sub-range, wherein the second sampling rate is twice the first sampling rate. The processor is also configured to apply the second convolutional layer to the intermediate set generated in the first convolutional layer at the first sampling rate when the object is within the second sub-range. The processor is also configured to define a third sub-range relative to the vehicle, wherein the third sub-range is smaller than the second sub-range; apply the first convolutional layer at a third sampling rate when the object is within the third sub-range, wherein the third sampling rate is twice the second sampling rate; apply the second convolutional layer to the intermediate set generated in the first convolutional layer at the second sampling rate; and apply the third convolutional layer to the output of the second convolutional layer at the first sampling rate. In another embodiment where the one or more convolutional layers include a first convolutional layer, the processor is further configured to define a first subrange relative to the vehicle and a second subrange relative to the vehicle, wherein the second subrange is smaller than the first subrange; and perform one of applying the first convolutional layer to the detection set once within the first subrange and applying the first convolutional layer twice within the second subrange, wherein the sampling rate of the first convolutional layer is the same within the first subrange and within the second subrange. The processor is also configured to define the first subrange and the second subrange, and define a first weight for the first subrange and a second weight for the second subrange to select a convolution path, wherein at a radial boundary between the first subrange and the second subrange, the first weight and the second weight decrease linearly with radial distance, wherein the sum of the first weight and the second weight is equal to 1.

[0007] In another exemplary embodiment, a vehicle is disclosed. The vehicle includes a sensor and a processor. The sensor obtains a detection set associated with an object at a range from the vehicle. The processor is configured to select a convolutional path for the detection set based on the range of the object, wherein the convolutional path includes one or more convolutional layers, and the number of the one or more convolutional layers depends on the range of the object; apply the one or more convolutional layers of the selected convolutional path to the detection set to generate a filtered data set; and operate the vehicle relative to the object using the filtered data set.

[0008] In addition to one or more features described herein, the one or more convolutional layers include a first convolutional layer, and the sampling rate of the first convolutional layer increases as the range decreases. In one embodiment where the one or more convolutional layers include the first convolutional layer, the processor is further configured to define a first sub-range relative to the vehicle and a second sub-range relative to the vehicle, wherein the second sub-range is smaller than the first sub-range; apply the first convolutional layer at a first sampling rate when the object is within the first sub-range; and apply the first convolutional layer at a second sampling rate when the object is within the second sub-range, wherein the second sampling rate is twice the first sampling rate. The processor is also configured to apply the second convolutional layer to the intermediate set generated in the first convolutional layer at the first sampling rate when the object is within the second sub-range. In another embodiment where the one or more convolutional layers include the first convolutional layer, the processor is further configured to define a first sub-range relative to the vehicle and a second sub-range relative to the vehicle, wherein the second sub-range is smaller than the first sub-range; and perform one of applying the first convolutional layer once within the first sub-range and applying the first convolutional layer twice within the second sub-range, wherein the sampling rate of the first convolutional layer is the same within the first sub-range and within the second sub-range. The processor is also configured to define a first sub-range and a second sub-range, and to define a first weight for the first sub-range and a second weight for the second sub-range to select a convolution path, wherein at a radial boundary between the first sub-range and the second sub-range, the first weight and the second weight decrease linearly with the radial distance, and wherein the sum of the first weight and the second weight is equal to 1.

[0009] The above features and advantages and other features and advantages of the present disclosure will become apparent from the following detailed description read in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Other features, advantages and details are given by way of example only in the following detailed description, which is made with reference to the accompanying drawings, in which:

[0011] Figure 1 An autonomous vehicle in an exemplary embodiment is shown;

[0012] Figure 2 A schematic diagram showing the effect of the distance between a vehicle and an object on the resolution of the object;

[0013] Figure 3 A block diagram showing a method for extracting features from an object using a set of range-dependent convolutional layers;

[0014] Figure 4 A block diagram showing an alternative method for extracting features from objects using a set of range-dependent convolutional layers; and

[0015] Figure 5 Shows Figure 3 and Figure 4 Schematic diagram of the range-dependent weighting of the convolutional path shown. DETAILED DESCRIPTION

[0016] The following description is merely exemplary in nature and is not intended to limit the present disclosure, its application or uses. It should be understood that throughout the drawings, corresponding reference numerals indicate similar or corresponding components and features.

[0017] According to an exemplary embodiment, Figure 1 An autonomous vehicle 10 is shown. In an exemplary embodiment, the autonomous vehicle 10 is a so-called Level 4 or Level 5 automation system. A Level 4 system represents "high automation" and refers to the autonomous driving system performing all aspects of the dynamic driving task in the driving mode, even if the driver does not respond appropriately to the intervention request. A Level 5 system represents "full automation" and refers to the autonomous driving system performing all aspects of the dynamic driving task full time under all road and environmental conditions that can be managed by the driver. It should be understood that the systems and methods disclosed herein can also be used for autonomous vehicles operating at any level from one to five.

[0018] The autonomous 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 sensor system 30, an actuator system 32, and a controller 34. The navigation system 20 determines a trajectory plan for autonomous driving of the autonomous vehicle 10. The propulsion system 22 provides power to generate power for the autonomous vehicle 10, and may include an internal combustion engine, an electric motor (e.g., a traction motor), and / or a fuel cell propulsion system in multiple embodiments. The transmission system 24 is configured to transmit power from the propulsion system 22 to two or more wheels 16 of the autonomous vehicle 10 at a selectable speed ratio. The steering system 26 affects the position of the two or more wheels 16. Although the steering system 26 is described as including a steering wheel 27 for exemplary purposes, in some embodiments contemplated within the scope of the present disclosure, the steering system 26 may not include a steering wheel 27. The braking system 28 is configured to provide braking torque to the two or more wheels 16.

[0019] The sensor system 30 includes a three-dimensional sensor 40 that senses objects in the external environment of the autonomous vehicle 10 and provides various parameters of the objects that can be used to locate the position and relative velocity of various remote vehicles in the autonomous vehicle environment. Such parameters can be provided to the navigation system 20. The three-dimensional sensor can be a radar or a lidar, or any other sensor that works using polar coordinates or spherical coordinates. In operation, the transmitter 42 of the three-dimensional sensor 40 sends a source signal 48, which is reflected back at the autonomous vehicle 10 by an object 50 within the field of view of the three-dimensional sensor 40 as one or more reflected echo signals 52 (also referred to herein as detections), which are received at the receiver 44. The one or more echo signals 52 can be used to determine various parameters of one or more objects 50, such as the range of the object, the Doppler frequency or relative radial velocity of the object, and the orientation. The sensor system 30 can also include additional sensors for identifying road features, etc., such as digital cameras.

[0020] The navigation system 20 constructs a trajectory for the autonomous vehicle 10 based on the parameters from the three-dimensional sensor 40 and any other relevant parameters. The controller 34 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 to navigate the autonomous vehicle 10 relative to the object 50.

[0021] The controller 34 includes a processor 36 and a computer-readable storage device or storage medium 38. The computer-readable storage medium includes a program or instruction 39 that, when executed by the processor 36, operates the autonomous vehicle based at least on the parameters from the three-dimensional sensor and other relevant data. The storage medium 38 may also include a program or instruction 39 that, when executed by the processor 36, determines the state of the object 50 to allow the autonomous vehicle to navigate relative to the object.

[0022] Figure 2 Schematic diagram 200 showing the effect of vehicle-to-object distance on the resolution of the object. A source signal 205 emanates from a three-dimensional sensor 40. As is well known, the strength of the source signal 205 follows the inverse square law. A first agent 202 and a second agent 204 are shown within the range of the three-dimensional sensor 40. The first agent 202 is located farther from the three-dimensional sensor 40 than the second agent 204. Due to the nature of the source signal 205, a set of returned detections are received along the azimuth of the field of view of the three-dimensional sensor 40. These detections are recorded in the azimuth bins of the three-dimensional sensor 40.

[0023] An enlarged version of the first proxy 202 shows the detections 206 from the first proxy 202. An enlarged version of the second proxy 204 shows the detections 210 and 212 from the second proxy 204. Since the second proxy 204 is closer to the 3D sensor 40, the number of detections from the second proxy 204 is greater than the number of detections from the first proxy 202. The detections from the second proxy 204 are alternately divided into an "even" group including the detections 210 and an "odd" group including the detections 212. The detections 210 in the "even" group have the same shape as the detections 206 from the first proxy 202. By selecting every other detection, one group (e.g., the "odd" group) can be compared with the detections 206.

[0024] The method disclosed herein receives detections from the vehicle's surroundings into azimuth bins and applies a series of convolutional layers to the azimuth bins to extract features of the objects. These convolutional layers are applied sequentially, with the output of one convolutional layer serving as the input to the next convolutional layer in the sequence. Each convolutional layer applies the filters of that layer to its input. The number of convolutional layers applied to the azimuth bins varies with the range of the object, with the number of convolutional layers increasing as the range decreases. In multiple embodiments, the range can be divided into multiple sub-ranges. Once the sub-range in which the object is located is determined, a selected convolutional path containing an appropriate number of convolutional filters can be applied to the set of detections. In one embodiment, the sampling rate of the convolutional layers can increase as the range decreases. In another embodiment, the number of times the convolutional layers are applied to the set of detections can increase as the range decreases.

[0025] Figure 3 FIG. 300 is a block diagram showing a method of extracting features from an object using a set of range-dependent convolutional layers. The method includes applying one or more convolutional filters to a set of detection points received from the object. The number of convolutional layers depends on the range of the object and increases as the range decreases. For a selected convolutional layer, the sampling rate increases as the range decreases.

[0026] The range of the 3D sensor 40 can be divided into multiple sub-ranges. In Figure 3 , the range 301 of the 3D sensor 40 is divided into a first sub-range 302, a second sub-range 304, and a third sub-range 306. The radial boundaries of these sub-ranges are identified by the radial distances R1, R2, and R max , where R1 < R2 < R max . The first sub-range 302 is defined by the radial distance between R2 and R max (i.e., R2 < r <= R max ). The second sub-range 304 is defined by the radial distance between R1 and R2 (i.e., R1 < r <= R2). The third sub-range 306 is defined by the radial distance between 0 and R1 (i.e., 0 < r <= R1).

[0027] Due to the inverse square law, the resolution of the detection decreases with the distance from the three-dimensional sensor 40. The detections received from the first sub-range 302 constitute a low-resolution detection set 303. The detections received from the second sub-range 304 constitute a medium-resolution detection set 305. The detections received from the third sub-range 306 constitute a high-resolution detection set 307. Assuming that the detections are based on the same source signal, the density of the high-resolution detection set 307 is greater than the density of the medium-resolution detection set 305, and the density of the medium-resolution detection set 305 is greater than the density of the low-resolution detection set 303.

[0028] The low-resolution detection set 303 is processed along a first convolutional path 310 including a first convolutional layer 312. The first convolutional layer 312 includes a first filter 314 applied to the low-resolution detection set 303 at a first sampling rate. For exemplary purposes, the first filter 314 is shown as having three orientation-dependent coefficients 316, each of which is applied to its appropriate orientation bin during convolution. Applying the first convolutional layer 312 to the low-resolution detection set 303 produces a filtered data set from which features can be extracted.

[0029] The medium resolution detection set 305 is processed along the second convolution path 320. The second convolution path 320 includes a first convolution layer 312 (operating at a second sampling rate twice the first sampling rate) and a second convolution layer 326 operating at the first sampling rate. The first convolution layer 312 samples the medium resolution detection set 305 using a second filter 322 (sampling at the second sampling rate). The second filter 322 is modified from the first filter 314 by adding a zero coefficient value 324 at every other azimuth bin. Applying the second filter 322 selects every other azimuth bin from the medium resolution detection set 305 and produces a first convolution set. The first convolution set is then processed by a second convolution layer 326 operating at the first sampling rate. The second convolution layer 326 applies a third filter 328 to the first convolution set (applied at the first sampling rate) to produce a filtered data set from which features can be extracted. It should be understood that each convolution block 312, 326, 336 can be a single convolution layer or a cascade of multiple convolution layers.

[0030] The high resolution detection set 307 is processed along a third convolutional path 330. The third convolutional path 330 includes a first convolutional layer 312 (operating at a third sampling rate that is four times the first sampling rate), a second convolutional layer 326 (operating at the second sampling rate), and a third convolutional layer 336 (operating at the first sampling rate).

[0031] The first convolution layer 312 samples the high resolution detection set 307 (sampling at a third sampling rate) using a fourth filter 332 to produce a first convolution set. The fourth filter 332 is modified from the second filter 322 by adding zero coefficient values ​​between each bin of the second filter 322. The first convolution set produced by the fourth filter 332 is then processed by the second convolution layer 326 using a fifth filter 334 at a second sampling rate to produce a second convolution set. The fifth filter 334 is modified from the third filter 328 by adding zero coefficient values ​​between each bin of the third filter 328. The second convolution set is then processed by the third convolution layer 336 (operating at the first sampling rate) using a sixth filter 338 to produce a filtered data set from which features can be extracted.

[0032] Changing operations from one convolution path to another depending on the range of the object maintains spatial invariance of the object across the range and field of view of the three-dimensional sensor 40. The operations are performed in the natural polar coordinates used by the three-dimensional sensor 40 to prevent interpolation losses. The hierarchical nature of the convolution system reduces the number of filters that need to be applied to the detection.

[0033] It should be understood that range 301 may be divided into more than three sub-ranges and may be divided into more than three sub-ranges according to reference Figure 3 The method described is to establish additional convolution paths to filter these additional sub-ranges.

[0034] Figure 4 A block diagram 400 of an alternative method for extracting features from an object using a set of range-dependent convolutional layers is shown. In this alternative method, the number of times the convolutional layers are applied depends on the range of the object and increases as the range decreases. The sampling range of the convolutional layers remains the same as the range. However, the sampling rate of subsequent convolutional layers increases.

[0035] Block diagram 400 shows a first convolution path 410, a second convolution path 420, and a third convolution path 430. The first convolution path 410 operates on the low-resolution detection set 303, the second convolution path 420 operates on the medium-resolution detection set 305, and the third convolution path 430 operates on the high-resolution detection set 307.

[0036] The first convolutional path 410 includes a first convolutional layer 312 that samples the low-resolution detection set 303 at a first sampling rate (fs / 4) to produce an output x1 from which features of the object can be extracted.

[0037] The second convolution path 420 includes the first convolution layer 312 and the second convolution layer 326. The first convolution layer 312 is applied to the medium resolution detection set 305 twice to obtain two intermediate sets x2 and x3. The intermediate set x2 is based on the odd-numbered azimuth bins 305a of the medium resolution detection set 305, while the intermediate set x3 is based on the even-numbered azimuth bins 305b of the medium resolution detection set 305.

[0038] The intermediate set x2 and the intermediate set x3 are then combined together at the combination layer 402 to form a combined data set y1. Combining the intermediate sets may include summing corresponding pixels of the intermediate sets x2 and x3. Alternatively, combining the outputs may include averaging the values ​​of the corresponding pixels. The second convolutional layer 326 is then applied to the combined data set y1 at a second sampling rate (fs / 2) to produce an output z1. Features of the object may then be extracted from the output z1.

[0039] The third convolution path 430 includes a first convolution layer 312, a second convolution layer 326, and a third convolution layer 336. The first convolution layer 312 is applied to the high-resolution detection set 307 four times at a first sampling rate. The first application of the first convolution layer 312 samples a first bin set 307a including every fourth bin starting from the first orientation bin. The second application of the first convolution layer 312 samples a second bin set 307b including every fourth bin starting from the second orientation bin. The third application of the first convolution layer 312 samples a third bin set 307c including every fourth bin starting from the third orientation bin. The fourth application of the first convolution layer 312 samples a fourth bin set 307d including every fourth bin starting from the fourth orientation bin.

[0040] The first, second, third, and fourth applications of the first convolutional layer 312 produce intermediate sets x4, x5, x6, and x7, respectively. Intermediate sets x4 and x6 are combined at the combination layer 404 to produce the combined set y2, and intermediate sets x5 and x7 are combined at the combination layer 406 to produce the combined set y3. The second convolutional layer 326 is applied to the combined set y2 at the second sampling rate to produce the intermediate set z2, and is applied to the combined set y3 at the second sampling rate to produce the intermediate set z3. Intermediate sets z2 and z3 are then combined at the combination layer 408 to obtain the combined set w. The third convolutional layer 336 is applied to the combined set w at a third sampling rate (fs) to produce a filtered data set from which features of the object can be extracted.

[0041] Figure 5 Shows Figure 3 and Figure 4Schematic diagram 500 of range-dependent weighting of convolution paths shown. The weighting system includes a first weight 502 applied to a first sub-range 302, a second weight 504 applied to a second sub-range 304, and a third weight 506 applied to a third range 306. The first weight 502, the second weight 504, and the third weight 506 are all normalized to be between a maximum value of 1 and a zero value. The weight represents the degree to which the convolution path is applied to the detection set to produce an output. The weight of the selected sub-range is at a maximum value over almost the entire radial extension of the sub-range, and decreases linearly with radial distance at the end of the sub-range. As an object passes from one sub-range to an adjacent sub-range, the weight of a convolution path may decrease linearly, while the weight of an adjacent sub-range may increase. For example, in a region 508 at or near the radial boundary R2 between the first sub-range 302 and the second sub-range 304, as the object leaves the first sub-range 302, the first weight 502 decreases linearly from a maximum value to a zero value, as shown by a straight line 510. In addition, in region 508, as the object enters the second sub-range 304, the second weight 504 increases linearly from a zero value to a maximum value, as shown by a straight line 512. In region 508, the sum of the first weight 502 and the second weight 504 is equal to one.

[0042] Although the disclosure above is described with reference to exemplary embodiments, it will be appreciated by those skilled in the art that various changes can be made without departing from the scope of the present disclosure, and that equivalent replacements can be made for the elements therein. In addition, various modifications can be made without departing from the basic scope of the present disclosure to adapt specific conditions or materials to the teachings of the present disclosure. Therefore, the present disclosure is not limited to the specific embodiments disclosed, but is intended to cover all embodiments falling within its scope.

Claims

1. A method of operating a vehicle, comprising: Obtaining a detection set associated with an object at a range from the vehicle, the range being determined by a sensor system of the vehicle; Selecting a convolutional path of the detection set according to the range of the object, wherein the convolutional path includes one or more convolutional layers, and the number of the one or more convolutional layers depends on the range of the object; processing the detection set of the corresponding resolution along a corresponding convolutional path including a corresponding convolutional layer, the corresponding convolutional layer applying a corresponding filter of the detection set of the corresponding resolution at a corresponding sampling rate to produce a filtered data set; and Maneuver the vehicle relative to the object using the filtered dataset.

2. The method of claim 1, wherein: The one or more convolutional layers include a first convolutional layer, and a sampling rate of the first convolutional layer increases as the range of the object decreases.

3. The method of claim 1, wherein: The one or more convolutional layers include a first convolutional layer, and the method also includes defining a first sub-range relative to the vehicle and a second sub-range relative to the vehicle, wherein the second sub-range is smaller than the first sub-range; applying the first convolutional layer at a first sampling rate when the object is within the first sub-range; and applying the first convolutional layer at a second sampling rate when the object is within the second sub-range, wherein the second sampling rate is twice the first sampling rate.

4. The method of claim 1, wherein: The one or more convolutional layers include a first convolutional layer, and the method further includes: defining a first sub-range relative to the vehicle and a second sub-range relative to the vehicle, wherein the second sub-range is smaller than the first sub-range; and applying the first convolutional layer to the detection set once within the first sub-range or applying the first convolutional layer twice within the second sub-range, wherein a sampling rate of the first convolutional layer is the same within the first sub-range and within the second sub-range.

5. The method as claimed in claim 1 further includes defining a first sub-range and a second sub-range, and defining a first weight for the first sub-range and a second weight for the second sub-range to select a convolution path, wherein at a radial boundary between the first sub-range and the second sub-range, the first weight and the second weight decrease linearly with the radial distance, and wherein the sum of the first weight and the second weight is equal to 1.

6. A system for operating a vehicle, comprising: a sensor for obtaining a detection set associated with an object at a range from the vehicle, the range being determined by the sensor; as well as A processor configured to: selecting a convolutional path of the detection set according to the range of the object, wherein the convolutional path includes one or more convolutional layers, and the number of the one or more convolutional layers depends on the range of the object, processing the detection set of the corresponding resolution along a corresponding convolutional path including a corresponding convolutional layer that applies a corresponding filter to the detection set of the corresponding resolution at a corresponding sampling rate to produce a filtered data set, and Maneuver the vehicle relative to the object using the filtered dataset.

7. The system of claim 6, wherein: The one or more convolutional layers include a first convolutional layer, and a sampling rate of the first convolutional layer increases as the range of the object decreases.

8. The system of claim 6, wherein: The one or more convolutional layers include a first convolutional layer, the processor further configured to define a first subrange relative to the vehicle and a second subrange relative to the vehicle, wherein the second subrange is smaller than the first subrange; apply the first convolutional layer at a first sampling rate when the object is within the first subrange; And applying the first convolutional layer at a second sampling rate when the object is within the second sub-range, wherein the second sampling rate is twice the first sampling rate.

9. The system of claim 6, wherein: The one or more convolutional layers include a first convolutional layer, and the processor is further configured to: define a first sub-range relative to the vehicle and a second sub-range relative to the vehicle, wherein the second sub-range is smaller than the first sub-range; and perform applying the first convolutional layer once to the detection set within the first sub-range or applying the first convolutional layer twice within the second sub-range, wherein a sampling rate of the first convolutional layer is the same within the first sub-range and within the second sub-range.

10. The system of claim 6, wherein: The processor is also configured to define a first sub-range and a second sub-range, and to define a first weight for the first sub-range and a second weight for the second sub-range to select a convolution path, wherein at a radial boundary between the first sub-range and the second sub-range, the first weight and the second weight decrease linearly with the radial distance, and wherein the sum of the first weight and the second weight is equal to 1.

Images