IP Library Granted Patent US 12,561,602
Granted Patent B2
US 12,561,602 · App. 17/002,650 · Granted Feb 24, 2026

Reinforcement learning based control of imitative policies for autonomous driving

Inventors: Zhangjie Cao (Stanford, CA); Erdem Biyik (Stanford, CA); Woodrow Zhouyuan Wang (Stanford, CA); Allan Raventos (San Francisco, CA); Adrien Gaidon (San Jose, CA); Guy Rosman (Newton, MA); Dorsa Sadigh (Stanford, CA)
Assignees: TOYOTA RESEARCH INSTITUTE, INC.; THE BROAD OF TRUSTEES OF THE LELAND STANFORD JUNIOR UNIVERSITY
G06N20/00B60W60/001B60W60/0011G06N3/092G06N7/01
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,561,602
App. No.
17/002,650
Granted
Feb 24, 2026
Kind
B2
Abstract

A method for controlling an ego agent includes periodically receiving policy information comprising a spatial environment observation and a current state of the ego agent. The method also includes selecting, for each received policy information, a low-level policy from a number of low-level policies. The low-level policy may be selected based on a high-level policy. The method further includes controlling an action of the ego agent based on the selected low-level policy.

Claims (29)

1 . A method for selecting a driving mode by a semi-autonomous vehicle, comprising:

periodically receiving policy information comprising a respective position and a respective velocity of each agent within a given distance from the semi-autonomous vehicle and a current position and a current velocity of the semi-autonomous vehicle within an environment;

selecting, in accordance with a high-level policy, a driving mode of a group of driving modes that maximizes a reward from a reward function that is based on a conditional observation probability of the policy information, the reward function accounting for resource use, a time to a destination, and collision avoidance, the group of driving modes including at least an aggressive driving mode and a defensive driving mode, the conditional observation probability modeling uncertainty due to sensor noise, the conditional observation probability defining a probability of observing the policy information given an unobserved behavior of one or more agents in the environment;

selecting an action from a plurality of actions associated with the driving mode, each driving mode of the group of driving modes associated with a respective low-level policy from a plurality of low-level policies trained via imitation learning, the imitation learning comprising supervised training of a policy model to directly map observed states to expert-labeled actions from mode-specific demonstrations provided by a group of human operators, each one of the group of human operators manually operating a training vehicle in accordance with one or more driving modes of the group of driving modes, the action including a throttle and a steering command that adjusts one or both of a current velocity or a current driving direction of the semi-autonomous vehicle, the action being selected in accordance with the reward function; and

overriding instructions from the human operator of the vehicle, such that the vehicle performs the action.

2 . The method of claim 1 , further comprising training, via reinforcement learning, a high-level policy model to learn the high-level policy based on each of the plurality of low-level policies.

3 . The method of claim 2 , in which training via the reinforcement learning comprises learning to maximize the reward function based on an action determined from a partially observable Markov decision process.

4 . An apparatus for selecting a driving mode for a semi-autonomous vehicle, comprising:

at least one processor;

at least one memory coupled with the at least one processor; and

instructions stored in the memory and operable, when executed by the at least one processor, to cause the apparatus to:

periodically receive policy information comprising a respective position and a respective velocity of each agent within a given distance from the semi-autonomous vehicle and a current position and a current velocity of the semi-autonomous vehicle within an environment;

select, in accordance with a high-level policy, a driving mode of a group of driving modes that maximizes a reward from a reward function that is based on a conditional observation probability of the policy information, the reward function accounting for resource use, a time to a destination, and collision avoidance, the group of driving modes including at least an aggressive driving mode and a defensive driving mode, the conditional observation probability modeling uncertainty due to sensor noise, the conditional observation probability defining a probability of observing the policy information given an unobserved behavior of one or more agents in the environment;

select an action from a plurality of actions associated with the driving mode, each driving mode of the group of driving modes associated with a respective low-level policy from a plurality of low-level policies trained via imitation learning, the imitation learning comprising supervised training of a policy model to directly map observed states to expert-labeled actions from mode-specific demonstrations provided by a group of human operators, each one of the group of human operators manually operating a training vehicle in accordance with one or more driving modes of the group of driving modes, the action including a throttle and a steering command that adjusts one or both of a current velocity or a current driving direction of the semi-autonomous vehicle, the action being selected in accordance with the reward function; and

override instructions from the human operator of the vehicle, such that the vehicle performs the action.

5 . The apparatus of claim 4 , in which the instructions further cause the apparatus to train, via reinforcement learning, a high-level policy model to learn the high-level policy based on each of the plurality of low-level policies.

6 . The apparatus of claim 5 , in which the instructions further cause the apparatus to train via the reinforcement learning comprises learning to maximize the reward function based on an action determined from a partially observable Markov decision process.

7 . A non-transitory computer-readable medium having program code recorded thereon for selecting a driving mode for a vehicle, the program code executed by at least one processor and comprising:

program code to periodically receive policy information comprising a respective position and a respective velocity of each agent within a given distance from the semi-autonomous vehicle and a current position and a current velocity of the semi-autonomous vehicle within an environment;

program code to select, in accordance with a high-level policy, a driving mode of a group of driving modes that maximizes a reward from a reward function that is based on a conditional observation probability of the policy information, the reward function accounting for resource use, a time to a destination, and collision avoidance, the group of driving modes including at least an aggressive driving mode and a defensive driving mode, the conditional observation probability modeling uncertainty due to sensor noise, the conditional observation probability defining a probability of observing the policy information given an unobserved behavior of one or more agents in the environment;

program code to select an action from a plurality of actions associated with the driving mode, each driving mode of the group of driving modes associated with a respective low-level policy from a plurality of low-level policies trained via imitation learning, the imitation learning comprising supervised training of a policy model to directly map observed states to expert-labeled actions from mode-specific demonstrations provided by a group of human operators, each one of the group of human operators manually operating a training vehicle in accordance with one or more driving modes of the group of driving modes, the action including a throttle and a steering command that adjusts one or both of a current velocity or a current driving direction of the semi-autonomous vehicle, the action being selected in accordance with the reward function; and

program code to override instructions from the human operator of the vehicle, such that the vehicle performs the action.

8 . The non-transitory computer-readable medium of claim 7 , in which the program code further comprises program code to train, via reinforcement learning, a high-level policy model to learn the high-level policy based on each of the plurality of low-level policies.

9 . The non-transitory computer-readable medium of claim 8 , in which the program code to train via the reinforcement learning comprises program code to learn to maximize the reward function based on an action determined from a partially observable Markov decision process.

10 . The method of claim 1 , in which the reward functions maximizes a return based on minimizing a travel time to a destination and maximizing safety of the vehicle.

11 . The apparatus of claim 4 , in which the reward functions maximizes a return based on minimizing a travel time to a destination and maximizing safety of the vehicle.

12 . The method of claim 1 , in which the high-level policy is periodically selected in response to a trigger associated with the policy information.

13 . The apparatus of claim 4 , in which the high-level policy is periodically selected in response to a trigger associated with the policy information.

14 . The non-transitory computer-readable medium of claim 7 , in which the high-level policy is periodically selected in response to a trigger associated with the policy information.

Assignments (3)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Apr 9, 2026
From: TOYOTA RESEARCH INSTITUTE, INC.
To: TOYOTA JIDOSHA KABUSHIKI KAISHA
Reel/Frame 074328/0606 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Nov 3, 2020
From: GAIDON, ADRIEN; ROSMAN, GUY; RAVENTOS, ALLAN
To: TOYOTA RESEARCH INSTITUTE, INC.
Reel/Frame 054260/0776 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Nov 3, 2020
From: CAO, ZHANGJIE; BIYIK, ERDEM; WANG, WOODROW ZHOUYUAN; SADIGH, DORSA
To: THE BOARD OF TRUSTEES OF THE LELAND STANFORD JUNIOR UNIVERSITY
Reel/Frame 054261/0398 →
Continuity (3)
Provisional Application 63052194 · Jul 15, 2020
Provisional Application 63040190 · Jun 17, 2020
Related Publication 20210398014A1 · Dec 23, 2021
References Cited (40)
US 10769525B2 · Redding · 2020 [cited by examiner]
US 20170297586A1 · Li · 2017 [cited by examiner]
US 20180113461A1 · Potnis · 2018 [cited by examiner]
US 20180170392A1 · Yang · 2018 [cited by examiner]
US 20190126472A1 · Tunyasuvunakool et al. · 2019 [cited by applicant]
US 20200089245A1 · Yadmellat · 2020 [cited by examiner]
US 20200104680A1 · Reed et al. · 2020 [cited by applicant]
US 20210001873A1 · Ingrody · 2021 [cited by examiner]
US 20210357782A1 · Graves · 2021 [cited by examiner]
Pineau, “A POMDP Tutorial”, 2013, European Workshop on Reinforcement Learning 2013, retrieved from https://www.cs.mcgill.ca/˜jpineau/talks/jpineau-dagstuhl13.pdf (Year: 2013). [cited by examiner]
Murphy, “A brief introduction to reinforcement learning”, 1998, retrieved from https://www.cs.ubc.ca/˜murphyk/Bayes/pomdp.html (Year: 1998). [cited by examiner]
Comanici et al., “Optimal policy switching algorithms for reinforcement learning”, 9th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2010), vol. 1-3, pp. 709-714 (Year: 2010). [cited by examiner]
Nair et al., “Overcoming Exploration in Reinforcement Learning with Demonstrations”, 2018, arXiv, v 1709.10089v2, pp. 1-8 (Year: 2018). [cited by examiner]
Qiao et al., “POMDP and Hierarchical Options MDP with Continuous Actions for Autonomous Driving at Intersections”, 2018, 2018 21st International Conference on Intelligent Transportation Systems (ITSC), vol. 21(2018), pp… [cited by examiner]
Wang et al., “Learning hierarchical behavior and motion planning for autonomous driving”, May 8, 2020, arXiv, v 2005.03863v1, pp. 1-8 (Year: 2020). [cited by examiner]
Daniel et al., “Probabilistic inference for determining options in reinforcement learning”, 2016, Machine Learning, vol. 104.2-3, pp. 337-357 (Year: 2016). [cited by examiner]
Chen et al., “Attention-based hierarchical deep reinforcement learning for lane change behaviors in autonomous Driving”, 2019, 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), vol. 2019, … [cited by examiner]
Hartmann et al., “Deep Reinforcement Learning for Time Optimal Velocity Control using Prior Knowledge”, Mar. 12, 2019, arXiv, v2, pp. 1-7 (Year: 2019). [cited by examiner]
Tingwall, “We Crash Four Cars Repeatedly to Test the Latest Automatic Braking Safety Systems”, 2018, retrieved from https://www.caranddriver.com/features/a24511826/safety-features-automatic-braking-system-tested-explain… [cited by examiner]
Duan et al., “Hierarchical reinforcement learning for self-driving decision-making without reliance on labelled driving data”, Feb. 11, 2020, Special Issue: AI Applications to Intelligent Vehicles for Advancing Intellig… [cited by examiner]
Zhang et al., “Hierarchical Reinforcement Learning for Multi-agent MOBA Game”, 2019, arXiv, v6, pp. 1-7 (Year: 2019). [cited by examiner]
Frans et al., “Meta Learning Shared Hierarchies”, 2017, arXiv, v1, pp. 1-11 (Year: 2017). [cited by examiner]
Abbel et al., “Apprenticeship learning via inverse reinforcement learning”, 2004, Proceedings of the twenty-first international conference on Machine learning, vol. 2004, pp. 1-8 (Year: 2004). [cited by examiner]
Zhu et al., “Human-like autonomous car-following model with deep reinforcement learning”, 2018, Transportation Research Part C, vol. 97 (2018), pp. 348-368 (Year: 2018). [cited by examiner]
Liaw et al., “Composing Meta-Policies for Autonomous Driving Using Hierarchical Deep Reinforcement Learning”, 2017, arXiv, v 1711.01503v1, pp. 1-8 (Year: 2017). [cited by examiner]
Richter et al., “Bayesian Learning for Safe High-Speed Navigation in Unknown Environments”, 2017, Springer Proceedings in Advanced Robotics, vol. 3, pp. 325-341 (Year: 2017). [cited by examiner]
Codevilla et al., “End-to-end Driving via Conditional Imitation Learning”, 2018, arXiv, v 1710.02410v2, pp. 1-8 (Year: 2018). [cited by examiner]
Abbeel et al., “Apprenticeship learning via inverse reinforcement learning”, 2004, ICML '04: Proceedings of the twenty-first international conference on Machine learning, vol. 2004, pp. 1-8 (Year: 2004). [cited by examiner]
Maurer et al., “Designing a Guardian Angel: Giving an Automated Vehicle the Possibility to Override its Driver”, 2018, AutomotiveUI '18: Proceedings of the 10th International Conference on Automotive User Interfaces and… [cited by examiner]
Jeong et al., “Surround Vehicle Motion Prediction Using LSTM-RNN for Motion Planning of Autonomous Vehicles at Multi-Lane Turn Intersections”, Jan. 13, 2020, IEEE Open Journal of Intelligent Transportation Systems, vol.… [cited by examiner]
Ohn-Bar et al., “Learning Situational Driving”, Jun. 13, 2020, 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), vol. 2020, pp. 11293-11302 (Year: 2020). [cited by examiner]
Kuefler et al., “Burn-In Demonstrations for Multi-Modal Imitation Learning”, 2017, arXiv, v1, pp. 1-8 (Year: 2017). [cited by examiner]
Silver et al., “Learning Autonomous Driving Styles and Maneuvers from Expert Demonstration”, 2013, Experimental Robotics: The 13th International Symposium on Experimental Robotics, vol. 13, pp. 371-386 (Year: 2013). [cited by examiner]
Kuderer et al., “Learning driving styles for autonomous vehicles from demonstration”, 2015, 2015 IEEE International Conference on Robotics and Automation (ICRA), vol. 2015, pp. 2641-2646 (Year: 2015). [cited by examiner]
Le, et al., “Hierarchical Imitation and Reinforcement Learning,” found at https://arxiv.org/abs/1803.00590, last revised Jun. 9, 2018. [cited by applicant]
Ngai, et al., “A Multiple-Goal Reinforcement Learning Method for Complex Vehicle Overtaking Maneuvers,” IEEE Transactions on Intelligent Transportation Systems, vol. 12, No. 2, pp. 509-522, Jun. 2011. [cited by applicant]
Chowdhuri, et al., “MultiNet: Multi-Modal Multi-Task Learning for Autonomous Driving,” found at https://arxiv.org/abs/1709.05581, last revised Jan. 14, 2019. [cited by applicant]
Li, et al., “InfoGAIL: Interpretable Imitation Learning from Visual Demonstrations,” found at https://arxiv.org/abs/1703.08840, last revised Nov. 14, 2017. [cited by applicant]
Dayan, et al., “Feudal Reinforcement Learning,” Advances in Neural Information Processing Systems, 1993, pp. 271-278. [cited by applicant]
Bhattacharyya, et al., “Multi-Agent Imitation Learning for Driving Simulation,” found at https://arxiv.org/abs/1803.01044, submitted Mar. 2, 2018. [cited by applicant]