Systems and methods for generating behavioral predictions in reaction to autonomous vehicle movement

1. A computer-implemented method for controlling an autonomous vehicle, the method comprising:

obtaining local scene data associated with an environment external to an autonomous vehicle, the local scene data comprising actor data for an actor in the environment external to the autonomous vehicle;

extracting, from the local scene data, one or more actor prediction parameters for the actor using a machine-learned parameter extraction model, wherein the actor prediction parameters comprises a latent space representation of at least a portion of the local scene data;

determining a plurality of candidate target motion trajectories for the autonomous vehicle;

generating, using a machine-learned prediction model, a plurality of reactive predictions for the actor based at least in part on the one or more actor prediction parameters, wherein the machine-learned prediction model is configured to be less computationally expensive than the machine-learned parameter extraction model, and wherein the plurality of reactive predictions are generated by:

for each respective candidate target motion trajectory of the plurality of candidate target motion trajectories:

inputting the latent space representation and the respective candidate target motion trajectory to the machine-learned prediction model;

generating, using the machine-learned prediction model, a respective reactive prediction for the actor based at least in part on the one or more actor prediction parameters and the respective candidate target motion trajectory, the respective reactive prediction comprising a probability representing a likelihood of the actor reacting in a particular manner to a movement of the autonomous vehicle based on the respective candidate target motion trajectory; and

operating the autonomous vehicle according to the respective candidate target motion trajectory.

2. The computer-implemented method of claim 1 , wherein the machine-learned prediction model executes a plurality of times for each execution of the machine-learned parameter extraction model.

3. The computer-implemented method of claim 2 , wherein operating the autonomous vehicle according to the respective candidate target motion trajectory comprises:

for the respective candidate target motion trajectory, assigning a candidate rating to the respective candidate target motion trajectory based at least in part on the respective reactive prediction for the actor;

selecting, based at least in part on the candidate rating for the respective candidate target motion trajectory, an optimal candidate target motion trajectory from the plurality of candidate target motion trajectories; and

operating the autonomous vehicle based on the optimal candidate target motion trajectory.

4. The computer-implemented method of claim 2 , further comprising:

generating, using the machine-learned prediction model, an interactive prediction for a first actor based at least in part on the one or more actor prediction parameters for the first actor, the respective candidate target motion trajectory, and a reactive prediction for a second actor, the interactive prediction comprising a probability representing a likelihood of the first actor reacting in a particular manner to movement of the autonomous vehicle based on the respective candidate target motion trajectory and behavior of the second actor based on the reactive prediction for the second actor.

5. The computer-implemented method of claim 1 , wherein, within a prescribed period of time, the machine-learned parameter extraction model executes once and the machine-learned prediction model executes a plurality of times.

6. The computer-implemented method of claim 1 , wherein the machine-learned prediction model predicts actor behavior in response to multiple candidate motions plans using feature data that is extracted once per evaluation cycle.

7. The computer-implemented method of claim 1 , wherein the machine-learned prediction model executes more quickly than the machine-learned parameter extraction model.

8. An autonomous vehicle control system for controlling an autonomous vehicle, the autonomous vehicle control system comprising:

one or more processors; and

one or more non-transitory computer-readable media storing instructions that are executable by the one or more processors to cause the autonomous vehicle control system to perform operations, the operations comprising:

obtaining local scene data associated with an environment external to an autonomous vehicle, the local scene data comprising actor data for an actor in the environment external to the autonomous vehicle;

extracting, from the local scene data, one or more actor prediction parameters for the actor using a machine-learned parameter extraction model, wherein the actor prediction parameters comprise a latent space representation of at least a portion of the local scene data;

determining a plurality of candidate target motion trajectories for the autonomous vehicle;

generating, using a machine-learned prediction model, a plurality of reactive predictions for the actor based at least in part on the one or more actor prediction parameters, wherein the machine-learned prediction model is configured to be less computationally expensive than the machine-learned parameter extraction model, and wherein the plurality of reactive predictions are generated by:

for each respective candidate target motion trajectory of the plurality of candidate target motion trajectories:

inputting the latent space representation and the respective candidate target motion trajectory to the machine-learned prediction model;

generating, using the machine-learned prediction model, a respective reactive prediction for the actor based at least in part on the one or more actor prediction parameters and the respective candidate target motion trajectory, the respective reactive prediction comprising a probability representing a likelihood of the actor reacting in a particular manner to a movement of the autonomous vehicle based on the respective candidate target motion trajectory; and

operating the autonomous vehicle according to the respective candidate target motion trajectory.

9. The autonomous vehicle control system of claim 8 , wherein the machine-learned prediction model executes a plurality of times for each execution of the machine-learned parameter extraction model.

10. The autonomous vehicle control system of claim 8 , wherein, within a prescribed period of time, the machine-learned parameter extraction model executes once and the machine-learned prediction model executes a plurality of times.

11. The autonomous vehicle control system of claim 8 , wherein the machine-learned prediction model predicts actor behavior in response to multiple candidate motions plans using feature data that is extracted once per evaluation cycle.

12. The autonomous vehicle control system of claim 8 , wherein the machine-learned prediction model executes more quickly than the machine-learned parameter extraction model.

13. The autonomous vehicle control system of claim 9 , wherein operating the autonomous vehicle according to the respective candidate target motion trajectory comprises:

for the respective candidate target motion trajectory, assigning a candidate rating to the respective candidate target motion trajectory based at least in part on the respective reactive prediction for the actor;

selecting, based at least in part on the candidate rating for the respective candidate target motion trajectory, an optimal candidate target motion trajectory from the plurality of candidate target motion trajectories; and

operating the autonomous vehicle based on the optimal candidate target motion trajectory.

14. The autonomous vehicle control system of claim 9 , wherein the operations comprise:

generating, using the machine-learned prediction model, an interactive prediction for a first actor based at least in part on the one or more actor prediction parameters for the first actor, the respective candidate target motion trajectory, and a reactive prediction for a second actor, the interactive prediction comprising a probability representing a likelihood of the first actor reacting in a particular manner to movement of the autonomous vehicle based on the respective candidate target motion trajectory and behavior of the second actor based on the reactive prediction for the second actor.

15. One or more non-transitory computer-readable media storing instructions that are executable by one or more processors to cause an autonomous vehicle control system to perform operations, the operations comprising:

obtaining local scene data associated with an environment external to an autonomous vehicle, the local scene data comprising actor data for an actor in the environment external to the autonomous vehicle;

extracting, from the local scene data, one or more actor prediction parameters for the actor using a machine-learned parameter extraction model, wherein the actor prediction parameters comprise a latent space representation of at least a portion of the local scene data;

determining a plurality of candidate target motion trajectories for the autonomous vehicle;

generating, using a machine-learned prediction model, a plurality of reactive predictions for the actor based at least in part on the one or more actor prediction parameters, wherein the machine-learned prediction model is configured to be less computationally expensive than the machine-learned parameter extraction model, and wherein the plurality of reactive predictions are generated by:

for each respective candidate target motion trajectory of the plurality of candidate target motion trajectories:

inputting the latent space representation and the respective candidate target motion trajectory to the machine-learned prediction model;

generating, using the machine-learned prediction model, a respective reactive prediction for the actor based at least in part on the one or more actor prediction parameters and the respective candidate target motion trajectory, the respective reactive prediction comprising a probability representing a likelihood of the actor reacting in a particular manner to a movement of the autonomous vehicle based on the respective candidate target motion trajectory; and

operating the autonomous vehicle according to the respective candidate target motion trajectory.

16. The one or more non-transitory computer-readable media of claim 15 , wherein the machine-learned prediction model executes a plurality of times for each execution of the machine-learned parameter extraction model.

17. The one or more non-transitory computer-readable media of claim 15 , wherein, within a prescribed period of time, the machine-learned parameter extraction model executes once and the machine-learned prediction model executes a plurality of times.

18. The one or more non-transitory computer-readable media of claim 15 , wherein the machine-learned prediction model predicts actor behavior in response to multiple candidate motions plans using feature data that is extracted once per evaluation cycle.

19. The one or more non-transitory computer-readable media of claim 15 , wherein the machine-learned prediction model executes more quickly than the machine-learned parameter extraction model.

20. The one or more non-transitory computer-readable media of claim 16 , wherein the operations comprise:

generating, using the machine-learned prediction model, an interactive prediction for a first actor based at least in part on the one or more actor prediction parameters for the first actor, the respective candidate target motion trajectory, and a reactive prediction for a second actor, the interactive prediction comprising a probability representing a likelihood of the first actor reacting in a particular manner to movement of the autonomous vehicle based on the respective candidate target motion trajectory and behavior of the second actor based on the reactive prediction for the second actor.