Method and system for dynamically updating an environmental representation of an autonomous agent
The method for dynamically updating an environmental representation of an autonomous agent can include: receiving a set of inputs; generating an environmental representation; and updating the environmental representation. Additionally or alternatively, the method can include providing the environmental representation to a planning module and/or any other suitable processes. The method functions to generate and/or dynamically update an environmental representation to facilitate control of an autonomous agent.
1 . A method comprising:
with an object tracking system, determining an environmental representation based on a set of inputs from a sensor suite of a vehicle, the environmental representation comprising a plurality of object representations, each object representation labeled with a respective instance identifier;
scoring a hypothesis associated with at least a subset of object representations of the plurality;
refining the environmental representation based on the scored hypothesis, comprising relabeling the respective instance identifier of at least one object representation of the subset, wherein relabeling the respective instance identifier comprises relabeling an instance identifier, determined at a prior timestep, which represents distinctness of the at least one object representation of the subset; and
autonomously controlling the vehicle based on the refined environmental representation.
2 . The method of claim 1 , wherein the hypothesis is selected from a predetermined set of hypotheses comprising an object-merge hypothesis and an object-split hypothesis.
3 . The method of claim 2 , wherein the hypothesis is selected based on satisfaction of a proximity rule by the subset of object representations.
4 . The method of claim 1 , wherein the hypothesis is scored based on a set of object tracks, each object track of the set corresponding to the respective instance identifier of a respective object representation of the subset.
5 . The method of claim 4 , wherein scoring the hypothesis comprises evaluating a smoothness of an object trajectory of an object track.
6 . The method of claim 4 , further comprising: modifying the set of object tracks based on the relabeling of the respective instance identifier of the at least one object representation of the subset.
7 . The method of claim 6 , wherein the environmental representation is determined based on a set of historical environmental representations, wherein modifying the set of object tracks comprises relabeling object representations in the plurality of historical environmental representations.
8 . The method of claim 1 , wherein scoring the hypothesis comprises evaluating a set of individual metrics for each object representation of the subset.
9 . The method of claim 8 , wherein scoring the hypothesis further comprises evaluating global metrics for the plurality of object representations.
10 . The method of claim 1 , wherein scoring the hypothesis comprises evaluating a cost function for the hypothesis across a plurality of historical environmental representations.
11 . The method of claim 1 , wherein relabeling the respective instance identifier of the at least one object representation of the plurality of object representations comprises fusing a first and second object representation under a singular label.
12 . The method of claim 1 , wherein the set of inputs are received from the sensor suite at a first time, wherein the refined environmental representation is provided to an autonomous controller of the vehicle within a predetermined time interval after the first time, wherein the environmental representation is determined with the object tracking system during a first portion of the predetermined time interval, wherein refining the environmental representation occurs during a remainder of the predetermined time interval after the first portion of the predetermined time interval, wherein the vehicle is autonomously controlled based on the refined environmental representation after the predetermined time interval.
13 . The method of claim 1 , wherein relabeling the respective instance identifier is performed during operation of the vehicle.
14 . A system comprising:
an object tracking system configured to generate an environmental representation, based on a set of inputs from a sensor suite of a vehicle, within a time interval immediately following a measurement time associated with the set of inputs, wherein the environmental representation comprises a set of object representations;
a computing system configured to receive the environmental representation after a first portion of the time interval and, during a remainder of the time interval, repeatedly:
score hypotheses for the environmental representation, each hypothesis associated with a respective subset of the set of object representations; and
in response to a score of a hypothesis exceeding a threshold, relabel the respective subset of object representations according to the hypothesis to refine the environmental representation, wherein relabeling the respective subset of object representations comprises: relabeling an instance identifier, determined prior to the time interval, which represents distinctness of at least one object representation of the subset; and
a controller configured to operate the vehicle based on the refined environmental representation after the time interval.
15 . The system of claim 14 , wherein the threshold comprises a reference score of the environmental representation.
16 . The system of claim 14 , wherein hypotheses are selected from a predetermined set of hypotheses comprising an object-merge hypothesis and an object-split hypothesis.
17 . The system of claim 16 , wherein the hypotheses are selected based on satisfaction of a proximity rule by the subset of object representations.
18 . The system of claim 14 , wherein the hypotheses are scored based on a set of object tracks, each object track of the set corresponding to the instance identifier of a respective object representation of the subset.
19 . The system of claim 18 , wherein the hypotheses are scored based on a smoothness of object trajectories associated with the set of object tracks.
20 . The system of claim 18 , further comprising: modifying the set of object tracks based on the relabeling of the respective instance identifier of the at least one object representation of the subset.
21 . The system of claim 14 , wherein the hypotheses are scored by evaluating a cost function for each hypothesis across a plurality of historical environmental representations.