Method and system for dynamically updating an environmental representation of an autonomous agent

1. A method comprising:

determining a set of measurement inputs using a sensor suite of a vehicle;

using an object tracking system, generating an environmental representation based on the measurement inputs, the environmental representation comprising a set of labelled objects in an environment of the vehicle, wherein the labelled objects are labelled with an instance identifier and a classification;

determining a hypothesis for a subset of the labelled objects;

determining a score for the hypothesis based on a plurality of historical environmental representations associated with the subset of labelled objects;

based on the score, updating the environmental representation, comprising: relabeling the subset of objects based on the hypothesis, wherein relabeling the subset of objects based on the hypothesis comprises modifying the classification and instance identifier of at least one labelled object of the subset;

providing the updated environmental representation to a processor onboard the vehicle; and

autonomously controlling the vehicle using the processor and based on the updated environmental representation.

2. The method of claim 1 , wherein the plurality of historical environmental representations comprises object tracks corresponding to each labelled object of the subset, each object track comprising the object instance identifier and historical object motion features, wherein the method further comprises: modifying the object tracks based on the relabeled subset of objects.

3. The method of claim 2 , wherein modifying the set of object tracks comprises relabeling objects in the plurality of historical environmental representations based on the hypothesis.

4. The method of claim 1 , wherein determining the score comprises evaluating a combination of individual metrics for each labelled object of the subset and global metrics for the set of labelled objects.

5. The method of claim 1 , wherein determining the score comprises evaluating a cost function for the hypothesis across the plurality of historical environmental representations.

6. The method of claim 1 , wherein determining the score comprises evaluating a smoothness of an object trajectory for a labelled object based on the plurality of historical environmental representations.

7. The method of claim 1 , wherein the hypothesis is selected from a predetermined set of hypotheses.

8. The method of claim 7 , wherein the predetermined set of hypotheses comprises an object-merge hypothesis and an object-split hypothesis.

9. The method of claim 7 , wherein the hypothesis is selected based on satisfaction of a proximity rule by the subset of labelled objects.

10. The method of claim 1 , wherein the historical environmental representation comprises set of object tracks, each object track comprising the instance identifier of a respective labelled object and historical object motion features for the respective labelled object, wherein the object tracking system generates the environmental representation based on the set of object tracks.

11. The method of claim 1 , wherein relabeling the subset of objects comprises fusing a first and second labelled object corresponding to measurement inputs from a first and second group of sensors, respectively, under a unitary object label.

12. The method of claim 11 , wherein the labelled objects are generated independently for input measurements corresponding to the first and second groups of sensors of the sensor suite.

13. The method of claim 11 , wherein the first group of sensors comprises a time-of-flight sensor, wherein the second group of sensors comprises a camera.

14. The method of claim 1 , wherein the set of measurement inputs are captured at a first time, wherein the updated environmental representation is provided to the processor within a predetermined time interval after the first time.

15. The method of claim 14 , wherein the interval is less than one-tenth of a second.

16. A system for a vehicle, comprising:

a sensor suite onboard the vehicle and configured to generate a set of measurements, wherein the set of measurements corresponds to a measurement time;

a memory comprising a set of historical environmental representations;

an object tracking system configured to generate an environmental representation based on the measurements within a time interval immediately following the measurement time, the environmental representation comprising a set of labelled objects;

a processing system configured to receive the environmental representation after a first portion of the time interval and, during a remainder of the time interval, repeatedly:

determine a first score for the environmental representation based on the set of historical environmental representations;

determine a hypothesis for a subset of the labelled objects;

determine a second score for the hypothesis based on the set of historical environmental representations; and

in response to the second score exceeding the first score, update the environmental representation, comprising: relabeling the subset of objects according to the hypothesis; and

a controller configured to receive the updated environmental representation from the processing system and operate the vehicle based on the updated environmental representation after the time interval.

17. The system of claim 16 , wherein the set of historical environmental representations comprises an object track corresponding to each labelled object of the subset, each object track comprising an object instance identifier and historical object motion features, wherein the processor is further configured to: modify the object tracks based on the updated environmental representation.

18. The system of claim 16 , wherein relabeling the subset of objects comprises fusing a first and second labelled object corresponding to measurement inputs from a first and second group of sensors, respectively, under a unitary object label.

19. The system of claim 18 , wherein the first group of sensors comprises a time-of-flight sensor, wherein the second group of sensors comprises a camera.