Trajectory prediction

1. A system for trajectory prediction, comprising:

a processor and a memory storing instructions, which when executed by the processor cause the processor to perform:

receiving a light detection and ranging (LiDAR) image sequence including a set of LiDAR images;

generating a LiDAR map based on the LiDAR image sequence;

generating an interaction encoder result by feeding the LiDAR image sequence through an interaction encoder;

generating a feature extractor result by feeding the LiDAR map through a feature extractor;

generating a relation encoder result by feeding a past trajectory of a detected obstacle from the LiDAR image sequence from a trajectory predictor, the interaction encoder result, and the feature extractor result through a relation encoder, wherein the trajectory predictor feeds back the past trajectory to the relation encoder;

generating an intention estimation result by feeding the relation encoder result through an intention estimator;

generating a conditional generative model result by feeding the past trajectory of the detected obstacle, the intention estimation result, and a probability map through a conditional generative model encoder; and

generating a trajectory prediction for the obstacle by feeding the relation encoder result, the past trajectory of the detected obstacle, and the conditional generative model result through the trajectory predictor.

2. The system for trajectory prediction of claim 1 , wherein the interaction encoder result and the feature extractor result are summed and provided as an input to the relation encoder.

3. The system for trajectory prediction of claim 1 , wherein the feature extractor or the interaction encoder include one or more 2D convolutional layers.

4. The system for trajectory prediction of claim 1 , wherein the intention estimator includes one or more non-linear unit layers and one or more fully connected layers.

5. The system for trajectory prediction of claim 1 , wherein the conditional generative model encoder includes one or more 2D convolutional layers and one or more fully connected layers, and wherein one or more of the 2D convolutional layers are associated with a non-linear unit function.

6. The system for trajectory prediction of claim 1 , wherein the trajectory predictor includes one or more fully connected layers and one or more deconvolutional layers.

7. The system for trajectory prediction of claim 1 , wherein generating the LiDAR map based on the LiDAR image sequence includes removing one or more dynamic obstacles from the LiDAR image sequence.

8. The system for trajectory prediction of claim 1 , wherein the intention estimation result is a probability distribution of intentional goals associated with different zones within an environment.

9. The system for trajectory prediction of claim 8 , wherein the trajectory prediction is a probability distribution of trajectories associated with the intentional goals associated with the different zones within the environment.

10. The system for trajectory prediction of claim 8 , wherein a zone of the different zones is a portion of an intersection.

11. A system for trajectory prediction, comprising:

a light detection and ranging (LiDAR) sensor receiving a LiDAR image sequence including a set of LiDAR images;

a processor and a memory storing instructions, which when executed by the processor cause the processor to perform:

generating a LiDAR map based on the LiDAR image sequence;

generating an interaction encoder result by feeding the LiDAR image sequence through an interaction encoder;

generating a feature extractor result by feeding the LiDAR map through a feature extractor;

generating a relation encoder result by feeding a past trajectory of a detected obstacle from the LiDAR image sequence from a trajectory predictor, the interaction encoder result, and the feature extractor result through a relation encoder, wherein the trajectory predictor feeds back the past trajectory to the relation encoder;

generating an intention estimation result by feeding the relation encoder result through an intention estimator;

generating a conditional generative model result by feeding the past trajectory of the detected obstacle, the intention estimation result, and a probability map through a conditional generative model encoder; and

generating a trajectory prediction for the obstacle by feeding the relation encoder result, the past trajectory of the detected obstacle, and the conditional generative model result through the trajectory predictor; and

a vehicle system activating an associated vehicle system function based on the trajectory prediction for the obstacle.

12. The system for trajectory prediction of claim 1 , wherein the interaction encoder result and the feature extractor result are summed and provided as an input to the relation encoder.

13. The system for trajectory prediction of claim 1 , wherein the feature extractor or the interaction encoder include one or more 2D convolutional layers.

14. The system for trajectory prediction of claim 1 , wherein the intention estimator includes one or more non-linear unit layers and one or more fully connected layers.

15. The system for trajectory prediction of claim 1 , wherein the conditional generative model encoder includes one or more 2D convolutional layers and one or more fully connected layers, and wherein one or more of the 2D convolutional layers are associated with a non-linear unit function.

16. The system for trajectory prediction of claim 1 , wherein the trajectory predictor includes one or more fully connected layers and one or more deconvolutional layers.

17. A computer-implemented method for trajectory prediction, comprising:

receiving a light detection and ranging (LiDAR) image sequence including a set of LiDAR images;

generating a LiDAR map based on the LiDAR image sequence;

generating an interaction encoder result by feeding the LiDAR image sequence through an interaction encoder;

generating a feature extractor result by feeding the LiDAR map through a feature extractor;

generating a relation encoder result by feeding a past trajectory of a detected obstacle from the LiDAR image sequence from a trajectory predictor, the interaction encoder result, and the feature extractor result through a relation encoder, wherein the trajectory predictor feeds back the past trajectory to the relation encoder;

generating an intention estimation result by feeding the relation encoder result through an intention estimator;

generating a conditional generative model result by feeding the past trajectory of the detected obstacle, the intention estimation result, and a probability map through a conditional generative model encoder; and

generating a trajectory prediction for the obstacle by feeding the relation encoder result, the past trajectory of the detected obstacle, and the conditional generative model result through the trajectory predictor.

18. The computer-implemented method for trajectory prediction of claim 17 , comprising summing the interaction encoder result and the feature extractor result and providing the sum as an input to the relation encoder.

19. The computer-implemented method for trajectory prediction of claim 17 , wherein generating the LiDAR map based on the LiDAR image sequence includes removing one or more dynamic obstacles from the LiDAR image sequence.

20. The computer-implemented method for trajectory prediction of claim 17 , wherein the intention estimation result is a probability distribution of intentional goals associated with different zones within an environment.