Catherine Weaver
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Catherine Weaver has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Chen Tang
Department of Mechanical Engineering, University of California, Berkeley, CA, USA
3 shared publications - Masayoshi Tomizuka
UC Berkeley
3 shared publications - Wei Zhan
University of California, Berkeley, CA, USA
3 shared publications - Ce Hao
School of Computing, National University of Singapore, Singapore
2 shared publications - Kenta Kawamoto
Sony AI, Tokyo, Japan
2 shared publications - Marion Leibold
Chair of Automatic Control Engineering, Technical University of Munich, Munich, Germany
1 shared publication
Latest Publications
- 18 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Control Systems Technology
- Vol. 33 (4) , 1301-1316
Active Exploration in Iterative Gaussian Process Regression for Uncertainty Modeling in Autonomous Racing
Autonomous racing creates challenging control problems, but model predictive control (MPC) has made promising steps toward solving both the minimum lap-time problem and head-to-head racing. Yet, accurate models of the system are necessary for model-based control, including models of vehicle dynamics and opponent behavior. Both dynamics model error and opponent behavior can be modeled with Gaussian process (GP) regression. GP models can be updated iteratively from data collected using the controller, but the strength of the GP model depends on the diversity of the training data. We propose a novel active exploration mechanism for iterative GP regression that purposefully collects additional data at regions of higher uncertainty in the GP model. In the exploration, an MPC collects diverse data by balancing the racing objectives and the exploration criterion; then the GP is retrained. The process is repeated iteratively; in later iterations, the exploration is deactivated, and only the racing objectives are optimized. Thus, the MPC can achieve better performance by leveraging the improved GP model. We validate our approach in the highly realistic racing simulation platform Gran Turismo Sport of Sony Interactive Entertainment Inc for a minimum lap time challenge, and in numerical simulation of head-to-head. Our active exploration mechanism yields a significant improvement in the GP prediction accuracy compared to previous approaches and, thus, an improved racing performance.
- 18 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Control Systems Technology
- Vol. 33 (4) , 1301-1316
Active Exploration in Iterative Gaussian Process Regression for Uncertainty Modeling in Autonomous Racing
Autonomous racing creates challenging control problems, but model predictive control (MPC) has made promising steps toward solving both the minimum lap-time problem and head-to-head racing. Yet, accurate models of the system are necessary for model-based control, including models of vehicle dynamics and opponent behavior. Both dynamics model error and opponent behavior can be modeled with Gaussian process (GP) regression. GP models can be updated iteratively from data collected using the controller, but the strength of the GP model depends on the diversity of the training data. We propose a novel active exploration mechanism for iterative GP regression that purposefully collects additional data at regions of higher uncertainty in the GP model. In the exploration, an MPC collects diverse data by balancing the racing objectives and the exploration criterion; then the GP is retrained. The process is repeated iteratively; in later iterations, the exploration is deactivated, and only the racing objectives are optimized. Thus, the MPC can achieve better performance by leveraging the improved GP model. We validate our approach in the highly realistic racing simulation platform Gran Turismo Sport of Sony Interactive Entertainment Inc for a minimum lap time challenge, and in numerical simulation of head-to-head. Our active exploration mechanism yields a significant improvement in the GP prediction accuracy compared to previous approaches and, thus, an improved racing performance.
- 19 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 9 (8) , 7302-7309
BeTAIL: Behavior Transformer Adversarial Imitation Learning From Human Racing Gameplay
Autonomous racing poses a significant challenge for control, requiring planning minimum-time trajectories under uncertain dynamics and controlling vehicles at their handling limits. Current methods requiring hand-designed physical models or reward functions specific to each car or track. In contrast, imitation learning uses only expert demonstrations to learn a control policy. Imitated policies must model complex environment dynamics and human decision-making. Sequence modeling is highly effective in capturing intricate patterns of motion sequences but struggles to adapt to new environments or distribution shifts that are common in real-world robotics tasks. In contrast, Adversarial Imitation Learning (AIL) can mitigate this effect, but struggles with sample inefficiency and handling complex motion patterns. Thus, we propose BeTAIL: Behavior Transformer Adversarial Imitation Learning, which combines a Behavior Transformer (BeT) policy from human demonstrations with online AIL. BeTAIL adds an AIL residual policy to the BeT policy to model the sequential decision-making process of human experts and correct for out-of-distribution states or shifts in environment dynamics. We test BeTAIL on three challenges with expert-level demonstrations of real human gameplay in the high-fidelity racing game Gran Turismo Sport. Our proposed BeTAIL reduces environment interactions and improves racing performance and stability, even when the BeT is pretrained on different tracks than downstream learning. Videos and code available at: https://sites.google.com/berkeley.edu/BeTAIL/home .
- 19 June 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 9 (8) , 7302-7309
BeTAIL: Behavior Transformer Adversarial Imitation Learning From Human Racing Gameplay
Autonomous racing poses a significant challenge for control, requiring planning minimum-time trajectories under uncertain dynamics and controlling vehicles at their handling limits. Current methods requiring hand-designed physical models or reward functions specific to each car or track. In contrast, imitation learning uses only expert demonstrations to learn a control policy. Imitated policies must model complex environment dynamics and human decision-making. Sequence modeling is highly effective in capturing intricate patterns of motion sequences but struggles to adapt to new environments or distribution shifts that are common in real-world robotics tasks. In contrast, Adversarial Imitation Learning (AIL) can mitigate this effect, but struggles with sample inefficiency and handling complex motion patterns. Thus, we propose BeTAIL: Behavior Transformer Adversarial Imitation Learning, which combines a Behavior Transformer (BeT) policy from human demonstrations with online AIL. BeTAIL adds an AIL residual policy to the BeT policy to model the sequential decision-making process of human experts and correct for out-of-distribution states or shifts in environment dynamics. We test BeTAIL on three challenges with expert-level demonstrations of real human gameplay in the high-fidelity racing game Gran Turismo Sport. Our proposed BeTAIL reduces environment interactions and improves racing performance and stability, even when the BeT is pretrained on different tracks than downstream learning. Videos and code available at: https://sites.google.com/berkeley.edu/BeTAIL/home .
- 21 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 9 (4) , 3625-3632
Skill-Critic: Refining Learned Skills for Hierarchical Reinforcement Learning
Hierarchical reinforcement learning (RL) can accelerate long-horizon decision-making by temporally abstracting a policy into multiple levels. Promising results in sparse reward environments have been seen with skills , i.e. sequences of primitive actions. Typically, a skill latent space and policy are discovered from offline data. However, the resulting low-level policy can be unreliable due to low-coverage demonstrations or distribution shifts. As a solution, we propose the Skill-Critic algorithm to fine-tune the low-level policy in conjunction with high-level skill selection. Our Skill-Critic algorithm optimizes both the low-level and high-level policies; these policies are initialized and regularized by the latent space learned from offline demonstrations to guide the parallel policy optimization. We validate Skill-Critic in multiple sparse-reward RL environments, including a new sparse-reward autonomous racing task in Gran Turismo Sport. The experiments show that Skill-Critic's low-level policy fine-tuning and demonstration-guided regularization are essential for good performance.
- 21 February 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 9 (4) , 3625-3632
Skill-Critic: Refining Learned Skills for Hierarchical Reinforcement Learning
Hierarchical reinforcement learning (RL) can accelerate long-horizon decision-making by temporally abstracting a policy into multiple levels. Promising results in sparse reward environments have been seen with skills , i.e. sequences of primitive actions. Typically, a skill latent space and policy are discovered from offline data. However, the resulting low-level policy can be unreliable due to low-coverage demonstrations or distribution shifts. As a solution, we propose the Skill-Critic algorithm to fine-tune the low-level policy in conjunction with high-level skill selection. Our Skill-Critic algorithm optimizes both the low-level and high-level policies; these policies are initialized and regularized by the latent space learned from offline demonstrations to guide the parallel policy optimization. We validate Skill-Critic in multiple sparse-reward RL environments, including a new sparse-reward autonomous racing task in Gran Turismo Sport. The experiments show that Skill-Critic's low-level policy fine-tuning and demonstration-guided regularization are essential for good performance.