Reinforcement Learning for Legged Robots: Truncated Quantile Critics with Path Following Tracking
Le, Hoan Quang So, Peter Abdelrahman, Ahmed Chen, Lingyun Le Mesle, Valentin Swikir, Abdalla Haddadin, Sami
Le, Hoan Quang
So, Peter
Abdelrahman, Ahmed
Chen, Lingyun
Le Mesle, Valentin
Swikir, Abdalla
Haddadin, Sami
Supervisor
Department
Robotics
Embargo End Date
Type
Conference proceeding
Date
2025
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
Path tracking control is a critical task for legged robots, such as hexapods, requiring them to navigate complex environments toward a series of predefined goals. This paper presents a reinforcement learning (RL)-based framework for efficient path-tracking, integrating Truncated Quantile Critics (TQC) for local target-reaching with a high-level path-planning algorithm. By breaking down the global path into sequential waypoints, the framework enables adaptive navigation and autonomous locomotion. Real-world experiments using CricketBOT, a hexapod robot, demonstrate the system’s effectiveness in tracking control and minimising errors. The results highlight the potential of RL strategies for real-time deployment in complex environments.
Citation
H. Q. Le et al., “Reinforcement Learning for Legged Robots: Truncated Quantile Critics with Path Following Tracking,” Springer Proceedings in Advanced Robotics, vol. 36 SPAR, pp. 274–280, 2025, doi: 10.1007/978-3-031-89471-8_42/FIGURES/3.
Source
Springer Proceedings in Advanced Robotics
Conference
16th European Robotics Forum, ERF 2025
Keywords
hexapod robot, path tracking, reinforcement learning
Subjects
Source
16th European Robotics Forum, ERF 2025
Publisher
Springer Nature