Publication
Our paper, titled Learning quadruped locomotion policies using logical rules, was published in the proceedings of International Conference on Automated Planning and Scheduling (ICAPS) 2024.
This work demonstrates how logical rules can be effectively integrated with reinforcement learning to improve quadruped robot locomotion, combining symbolic reasoning with continuous control.
Research Overview
Quadruped locomotion is a challenging problem in robotics that requires coordinated control of multiple legs while maintaining balance and adapting to various terrains. Our research introduces a novel approach that leverages logical rules to guide the learning process, making it more efficient and interpretable.
Key Innovation
Traditional reinforcement learning approaches for locomotion often require extensive trial-and-error learning. By incorporating logical rules that encode domain knowledge about quadruped gaits and stability principles, our method achieves:
- Faster Learning: Reduced training time through guided exploration
- Better Stability: Improved balance and robustness during locomotion
- Interpretable Policies: Clear understanding of decision-making process
- Transferable Knowledge: Rules that generalize across different scenarios
Technical Contributions
- Integration of symbolic logical rules with continuous RL policies
- Novel reward shaping mechanism based on locomotion principles
- Demonstration on both simulated and real quadruped robots
- Comprehensive evaluation across various terrains and conditions
Applications and Impact
This research has important implications for:
- Autonomous quadruped robots in challenging environments
- Search and rescue operations
- Inspection and monitoring tasks
- Advancing the integration of symbolic and subsymbolic AI
Project Resources
Our comprehensive project website includes videos, code, and additional technical details to support reproducibility and further research.
Authors: David DeFazio, Yohei Hayamizu, and Shiqi Zhang
Venue: International Conference on Automated Planning and Scheduling (ICAPS) 2024
| Paper | Project Website |