received by the International Bureau on 24 April 2020 (24.04.2020)

Claims

1. A computer-implemented method of probabilistically predicting a goal of an external actor, the method comprising:

receiving, at a computer, sensor inputs for detecting and tracking the external actor; applying object tracking to the sensor inputs, in order track the external actor, and thereby determine an observed trace of the external actor over a time interval;

determining a set of available goals for the external actor;

for each of the available goals, determining an expected trajectory model over said time interval; and

comparing the observed trace of the external actor with the expected trajectory model for each of the available goals, to determine a likelihood of that goal.

2. The method of claim 1 , implemented in an autonomous ego vehicle, wherein a planner of the autonomous vehicle makes an autonomous driving decision in dependence on the likelihood of at least one of the available goals, the sensor inputs obtained using a sensor system of the autonomous vehicle.

3. The method of claim 1 or 2, wherein the expected trajectory model is a single predicted trajectory associated with that goal or a distribution of predicted trajectories associated with that goal.

4. The method of claim 3, wherein the expected trajectory model is a distribution comprising a conditional probability for each predicted trajectory T in a set of

predicted trajectories and the likelihood of that goal is used to estimate at least one predicted trajectory probability given the observed trace t.

5. The method of any preceding claim, wherein the expected trajectory model is determined for each goal based on a desired goal location of that goal.

6. The method of any preceding claim, wherein the expected trajectory model is determined by executing a generative model for each goal, the generative behaviour model having been trained to generate trajectories based on examples of real-world driving behaviour.

7. The method of claim 6, wherein the models are specific to a driving area in relation to which the method is applied.

8. The method of any preceding claim, wherein the expected trajectory model is determined by applying a sampling algorithm to sample a space of predicted trajectories, which is defined for each goal based on one or more parameters of that goal and one or more parameters of the external actor.

9. The method of any preceding claim, wherein the set of available goals is determined based on map data associated with the external actor.

10. The method of any preceding claim, wherein the expected trajectory model is determined based on one or more ego vehicle parameters to model the other actor’s response to ego vehicle behaviour.

11. The method of any preceding claim, wherein the observed trace is used to predict a best-available trajectory model for the goal, said comparison comprising comparing the best- available trajectory model with the expected trajectory model.

12. The method of claim 11 , wherein the observed trace is used to predict a current maneuver and/or a future maneuver of the external actor, the predicted current or future maneuver being used to determine the best-available trajectory model.

13. The method of claim 12, wherein a sequence of multiple maneuvers is determined for at least one goal, the best-available trajectory model being determined for that goal based on partial trajectory models respectively associated with the multiple maneuvers.

14. The method of claim 13, wherein each partial trajectory model comprises one or more target motion values, and one or more motion values of a future portion of the best-available trajectory model are determined by applying motion smoothing to the target motion values.

15. The method of any of claims 1 1 to 14_» wherein the expected trajectory model for each goal is a single expected trajectory for that goal, and the best-available trajectory model for each goal is a single best-available trajectory. 16. The method of claim 15 when dependent on claim 13, wherein the partial trajectory model for each maneuver is a most-likely partial trajectory for that maneuver.

17. The method of any of claims 11 to 16, wherein a defined cost function is applied to both the expected trajectory model and the best-available trajectory model for each goal, to determine respective costs of those trajectory models, wherein said comparison comprises comparing those costs.

18. The method of claim 17, wherein the cost function rewards reduced driving time whilst penalizing unsafe trajectories.

19. The method of claim 18, wherein the cost function also penalizes lack of comfort.

20. The method of any preceding claim, comprising the step of determining at least one predicted trajectory for the external actor after said time interval for at least one of said goals, and a probability of the predicted trajectory based on the likelihood of the goal.

21. The method of claim 20, wherein the expected trajectory model in said time interval is determined based on a location of the external actor at the start of the time interval, and the at least one predicted trajectory is determined based on a location of the external actor at the end of said time interval.

22. The method of claim 21 , wherein the at least one predicted trajectory is determined by applying a sampling algorithm to sample a space of predicted trajectories, which is defined for the goal based on one or more parameters of that goal and the new location of the external actor.

23. A computer-implemented method of probabilistically predicting a maneuver of an external actor, the method comprising:

receiving, at a computer, sensor inputs for detecting and tracking an external actor; applying object tracking to the sensor inputs, in order track the external actor, and thereby determine an observed trace of the external actor over a time interval;

determining a set of possible maneuvers for the external actor;

for each of the possible maneuvers, determining an expected trajectory model over said time interval; and

comparing the observed trace of the external actor with the expected trajectory model for each of the available maneuvers, to determine a likelihood of that maneuver,

24. The method of claim 23, implemented in an autonomous vehicle, wherein a planner of the autonomous vehicle makes an autonomous driving decision in dependence on the likelihood of at least one of the available maneuvers.

25. The method of claim 23 or 24, wherein the expected trajectory model is a single predicted trajectory associated with that maneuver or a distribution of predicted trajectories associated with that maneuver.

26. The method of claim 25, wherein the observed trace is compared with a most-likely trajectory of the distribution of predicted trajectories.

27. The method of any of claims 23 to 26, comprising the step of determining at least one predicted trajectory for the external actor after said time interval for at least one of said maneuvers, and a probability of the predicted trajectory based on the likelihood of the maneuvers,

28. The method of claim 27, wherein the expected trajectory model in said time interval is determined based on a location of the external actor at the start of the time interval, and the at least one predicted trajectory is determined based on a location of the external actor at the end of said time interval.

29. The method of claim 28, wherein the at least one predicted trajectory is determined by applying a sampling algorithm to sample a space of predicted trajectories, which is defined for the maneuver based on one or more parameters of that maneuver and the new location of the external actor.

30. A computer system comprising execution hardware configured to execute the method of any preceding claim.

31, A computer program comprising executable instructions configured, when executed, to implement any of the method of any of claims 1 to 29,

32. An autonomous vehicle computer system comprising

a prediction component configured to implement the method of any of claims 1 to 31 ; and

a planner configured to make autonomous driving decisions using outputs of the prediction components.

33. The autonomous vehicle computer system of claim 32, wherein the prediction component is configured to implement the method of any of claims 1 to 19 to provide a goal prediction for an external actor and the method of any of claims 20 to 23 to provide a maneuver prediction for the external actor,

34, The autonomous vehicle computer system of claim 33, wherein the manoeuvre prediction is used to make the goal prediction.

35, An autonomous vehicle comprising the autonomous vehicle computer system of claim 32, 33 or 34 and a drive mechanism coupled to the planner and responsive to control signals generated by the planner.