Method for maneuver prediction of traffic participant
A method and system for navigating traffic for an autonomously controllable vehicle operates by detecting at least one target vehicle within a traffic environment with at least one sensor disposed on a primary vehicle. The system obtains information with a plurality of sensors disposed on the primary vehicle of the traffic environment and the at least target vehicle and predicts a maneuver of the at least one target vehicle with a controller disposed on the primary vehicle for a future time based on the information obtained relating to the traffic environment and the at least one target vehicle. The system further generates a recommended path for the primary vehicle with the controller that avoids conflicts with a future position of the at least one target vehicle based on predicted maneuvers of the at least one target vehicle.
1. A method of navigating traffic for an autonomously controllable vehicle, the method comprising:
detecting at least one target vehicle within a traffic environment with at least one sensor disposed on a primary vehicle;
obtaining information with a plurality of sensors disposed on the primary vehicle of the traffic environment and the at least target vehicle;
predicting a maneuver of the at least one target vehicle with a controller disposed on the primary vehicle for a future time based on the information obtained relating to the traffic environment and the at least one target vehicle; and
generating a recommended path for the primary vehicle with the controller that avoids conflicts with a future position of the at least one target vehicle based on predicted maneuvers of the at least one target vehicle,
wherein obtaining information on the traffic environment includes generating a local region feature map of an image region corresponding to the at least one target vehicle from images captured by cameras disposed on the primary vehicle, and
wherein predicting the maneuver of the at least one target vehicle includes utilizing a recurrent neural network that receives information obtained about the traffic environment and the local region feature map to predict possible positions and maneuvers of the at least one target vehicle.
2. The method as recited in claim 1 , wherein the recurrent neural network predicts possible positions and maneuvers of the at least one target vehicle for a plurality of future times.
3. The method as recited in claim 1 , wherein the information relating to the traffic environment includes information relating to a spatial relationship between the primary vehicle and the at least one target vehicle.
4. The method as recited in claim 1 , wherein the information relating to the traffic environment includes information relating to a spatial relationship between the at least one target vehicle and other traffic participants.
5. The method as recited in claim 1 , wherein the information relating to the traffic environment includes information relating to a spatial relationship between the at least one target vehicle and boundaries of a roadway.
6. The method as recited in claim 1 , wherein the information relating to the traffic environment includes information obtained directly from the at least one target vehicle.
7. The method as recited in claim 1 , wherein the primary vehicle comprises a partially autonomously vehicle that generates operation of the primary vehicle functions to assist in driver operation.
8. The method as recited in claim 1 , wherein the primary vehicle comprises a fully autonomous vehicle that operates independently.
9. The method of claim 1 , wherein generating the local region feature map employs a fully convolutional network to extract the local region feature map.
10. A vehicle system navigation system for navigating an autonomously controllable vehicle within a traffic environment comprising:
a detection module for obtaining information on proximate traffic participants;
a local region feature map that is continually updated based on information from the detection module;
a recurrent neural network for generating a plurality of predicted possible future positions for at least one traffic participant at a plurality of future time steps based at least partly on the local region feature map; and
a vehicle controller for generating commands to alter a path of the vehicle based on the plurality of predicted future positions of the at least one traffic participants.
11. The system as recited in claim 10 , wherein the vehicle detection module comprises a camera that captures images of the at least one traffic participant, other stationary and non-stationary objects surrounding a vehicle.
12. The system as recited in claim 11 , including at least one sensor providing information indicative of vehicle operation within the traffic environment.
13. The system as recited in claim 11 , wherein the vehicle controller is configured to detect relationships between the at least one traffic participant and other traffic participants.
14. The system as recited in claim 13 , wherein the at least one traffic participant includes a first vehicle and the vehicle controller is configured to detect a position of the first vehicle within boundaries of a roadway.
15. The system as recited in claim 14 , wherein the vehicle controller detects actions directly from the first vehicle that are indicative of a future maneuver.
16. The system as recited in claim 10 , wherein the vehicle controller generates is part of a partially autonomous vehicle and generates operation of vehicle functions to assist in driver operation.
17. The system as recited in claim 10 , wherein the vehicle controller generates commands for primary vehicle comprises a fully autonomous vehicle that operates independently.
18. The system of claim 10 , further comprising a fully convolutional network which extracts the local region feature map as an output of the fully convolutional network based on the information from the detection module, the local region feature map being provided as input to the recurrent neural network.