Method and system for context-aware decision making of an autonomous agent
A system for context-aware decision making of an autonomous agent includes a computing system having a context selector and a map. A method for context-aware decision making of an autonomous agent includes receiving a set of inputs, determining a context associated with an autonomous agent based on the set of inputs, and optionally any or all of: labeling a map; selecting a learning module (context-specific learning module) based on the context; defining an action space based on the learning module; selecting an action from the action space; planning a trajectory based on the action S 260 ; and/or any other suitable processes.
1. A method comprising:
determining a plurality of contexts for an autonomous agent;
for each context of the plurality, generating a context-specific learning model using a context-specific training dataset associated with the context;
determining a context mapping for a fixed route of an autonomous agent, comprising mapping each region of the fixed route to a respective context of the plurality; and
providing, at a computing system of the autonomous agent, the context mapping and the context-specific learning models, wherein the computing system is configured to selectively operate the autonomous agent in each region of the fixed route based on the context-specific learning model corresponding to the respective context of the region.
2. The method of claim 1 , wherein each context-specific learning model comprises a reward function generated using a learning algorithm for the context-specific training dataset associated with the context.
3. The method of claim 1 , wherein generating each context-specific learning model comprises: training the context-specific learning model to transform an environmental representation input into an output using inverse reinforcement learning based on the context-specific training dataset.
4. The method of claim 3 , wherein the inverse reinforcement learning is based on human-derived behaviors.
5. The method of claim 4 , wherein each context-specific learning model comprises an optimal policy generated based on the human-derived behaviors for the context specific training dataset associated with the context.
6. The method of claim 1 , wherein each context-specific learning model transforms an environmental representation input into an output action space; wherein the computing system is configured to operate the autonomous agent based on the action space.
7. The method of claim 6 , wherein the computing system is configured to operate the autonomous agent based on the action space by:
determining an action within the action space;
determining a trajectory based on the action; and
controlling the autonomous agent based on the trajectory.
8. The method of claim 6 , wherein the output action space consists of a unitary action.
9. The method of claim 6 , wherein the action space is associated with a set of parameters having context-specific values.
10. The method of claim 9 , wherein the set of parameters comprises a creep distance.
11. The method of claim 6 , wherein the environmental representation comprises a set of detected dynamic objects.
12. The method of claim 1 , further comprising facilitating traversal of the autonomous agent along the fixed route by:
determining a first context for the autonomous agent according to the context mapping; and
based on the first context, operating the autonomous agent based on a first context-specific learning model associated with the first context.
13. The method of claim 12 , wherein operating the autonomous agent comprises determining a trajectory for the autonomous agent based on an output of the respective context-specific leaning model associated with the first context.
14. The method of claim 12 , wherein facilitating traversal of the autonomous agent along the fixed route further comprises:
determining a second context for the autonomous agent according to the context mapping; and
based on the second context, operating the autonomous agent based on a second context-specific learning model associated with the second context.
15. The method of claim 1 , further comprising: updating the respective context-specific learning models based on operation data collected during operation of the autonomous agent along the fixed route and the context mapping.
16. The method of claim 1 , wherein each context-specific learning model comprises a trained convolutional neural network (CNN).
17. The method of claim 1 , wherein each context-specific learning model is provided, at the computing system of the autonomous agent, within a series of connected machine learning models.
18. The method of claim 17 , wherein each context-specific learning model is an intermediate element of the series.
19. The method of claim 17 , wherein the context-specific training datasets used to generate each context-specific learning model comprise environmental representation inputs, wherein the environmental representation inputs are generated as outputs of a first machine learning model.
20. The method of claim 17 , wherein the computing system is configured to operate the autonomous agent by:
generating an output for a selected context-specific learning model;
based on the output, determining a trajectory of the autonomous agent using a second machine learning model; and
operating the autonomous agent based on the trajectory.