Collision-avoidance system for autonomous-capable vehicles
A collision-avoidance system for use with an autonomous-capable vehicle can continuously receive image frames captured of the roadway to determine drivable space in a forward direction of the vehicle. The system can determine, for each image frame, whether individual regions of the image frame depict drivable space. The system can do so using machine-learned image recognition algorithms such as convolutional neural networks generated using extensive training data. Using such techniques, the system can label regions of the image frames as corresponding to drivable space or non-drivable space. By analyzing the labeled image frames, the system can determine whether the vehicle is likely to impact a region of non-drivable space. And, in response to such a determination, the system can generate control signals that override other control systems or human operator input to control the brakes, the steering, or other sub-systems of the vehicle to avoid the collision.
1. A vehicle comprising:
multiple sensors including a camera for capturing images;
a collision-avoidance system being configured to perform steps including:
receiving a first image from the camera;
performing image analysis on at least a portion of the first image to determine respective probabilities that regions of the first image depict drivable space;
determining a plurality of regions of the first image as depicting non-drivable space based on the respective probabilities;
determining that the vehicle is likely to impact one of the plurality of regions of the first image labeled as depicting non-drivable space; and
in response to determining that the vehicle is likely to impact one of the plurality of regions of the first image determined as depicting non-drivable space, generating a collision-avoidance signal, the collision-avoidance signal being associated with an action that avoids a collision or minimizes the impact of the collision of the vehicle with the region of non-drivable space.
2. The vehicle of claim 1 , further comprising a vehicle control system to control the vehicle, the vehicle control system being configured to (i) receive sensor input from the multiple sensors, and (ii) generate, based on the sensor input, vehicle control signals.
3. The vehicle of claim 2 , wherein the collision-avoidance system operates independently of the vehicle control system.
4. The vehicle of claim 2 , wherein the collision-avoidance signal overrides at least one of the vehicle control signals generated by the vehicle control system.
5. The vehicle of claim 2 , wherein the collision-avoidance system and the vehicle control system utilize a shared set of hardware resources including a shared memory resource and at least one shared processor.
6. The vehicle of claim 2 , wherein the collision-avoidance system and the vehicle control system utilize separate sets of hardware resources.
7. The vehicle of claim 1 , wherein the image analysis is performed using a machine-learned model.
8. The vehicle of claim 7 , wherein the machine-learned model is a convolution neural network.
9. The vehicle of claim 7 , wherein the machine-learned model is generated using training data, including training images.
10. The vehicle of claim 9 , wherein training data further includes sensor data and wherein each training image is associated with a set of corresponding sensor data captured contemporaneously with the training image.
11. The vehicle of claim 10 , wherein the sensor data comprises LIDAR measurements.
12. The vehicle of claim 7 , wherein the machine-learned model is generated based on a supervising input received from an administrator, the supervising input identifying portions of a training image as representative of drivable space or non-drivable space.
13. The vehicle of claim 1 , wherein determining the plurality of regions of the first image as depicting non-drivable space comprises:
identifying the plurality of regions of the first image by comparing each of the respective probabilities with a threshold value; and
labeling each of the plurality of regions of the first image as depicting non-drivable space.
14. The vehicle of claim 1 , wherein the collision-avoidance signal causes the vehicle to activate one or more brakes of the vehicle.
15. The vehicle of claim 1 , wherein the collision-avoidance system is configured to further perform steps including:
receiving a second image from the camera; and
wherein determining the plurality of regions of the first image as depicting non-drivable space is based further on analyzing at least a portion of the second image.
16. A collision-avoidance system on a vehicle, the collision-avoidance system comprising circuitry configured to:
receive a first image from a camera;
perform image analysis on at least a portion of the first image to determine respective probabilities that regions of the first image depict drivable space;
determine a plurality of regions of the first image as depicting non-drivable space based on the respective probabilities;
determine that the vehicle is likely to impact one of the plurality of regions of the first image labeled as depicting non-drivable space; and
in response to determining that the vehicle is likely to impact one of the plurality of regions of the first image determined as depicting non-drivable space, generate a collision-avoidance signal, the collision-avoidance signal being associated with an action that avoids a collision or minimizes the impact of the collision of the vehicle with the region of non-drivable space.
17. The collision-avoidance system of claim 16 , wherein the image analysis is performed using a machine-learned model, and wherein the machine-learned model includes a convolution neural network.
18. The collision-avoidance system of claim 16 , wherein the collision-avoidance system operates independently of a vehicle control system to control the vehicle, the vehicle control system being configured to (i) receive sensor input from multiple sensors of the vehicle, and (ii) generate, based on the sensor input, vehicle control signals.
19. A computer-implemented method of operating a collision-avoidance system on a vehicle, comprising:
receiving a first image from a camera;
performing image analysis on at least a portion of the first image to determine respective probabilities that regions of the first image depict drivable space;
determining a plurality of regions of the first image as depicting non-drivable space based on the respective probabilities;
determining that the vehicle is likely to impact one of the plurality of regions of the first image labeled as depicting non-drivable space; and
in response to determining that the vehicle is likely to impact one of the plurality of regions of the first image determined as depicting non-drivable space, generating a collision-avoidance signal, the collision-avoidance signal being associated with an action that avoids a collision or minimizes the impact of the collision of the vehicle with the region of non-drivable space.
20. The computer-implemented method of claim 19 , wherein the image analysis is performed using a machine-learned model, and wherein the machine-learned model includes a convolution neural network.