Related Experiment Video
Updated: Jun 23, 2025
11:18
Quantifying Learning in Young Infants: Tracking Leg Actions During a Discovery-learning Task
Published on: June 1, 2015
10.7K
Adaptive Gait Acquisition through Learning Dynamic Stimulus Instinct of Bipedal Robot
Yuanxi Zhang1, Xuechao Chen1,2, Fei Meng1,2
1School of Mechatronical Engineering, Beijing Institute of Technology, Beijing 100081, China.
Biomimetics (Basel, Switzerland)
|June 26, 2024
Summary
This study introduces dynamic stimulus signals into reinforcement learning for bipedal robots, enabling adaptive 3D locomotion and disturbance resistance. The method ensures robust performance, even in challenging real-world scenarios like foot sliding.
Area of Science:
- Robotics
- Control Theory
- Machine Learning
Background:
- Standard bipedal gaits rely on fixed stimulus signals, limiting adaptability to perturbations.
- Robots require more dynamic gaits to effectively address imbalances and external disturbances.
Purpose of the Study:
- To introduce dynamic stimulus signals within a reinforcement learning (RL) framework for bipedal locomotion.
- To enable robots to achieve adaptive 3D locomotion and disturbance resistance without explicit model-based gaits.
Main Methods:
- Integrating dynamic stimulus signals with a bipedal locomotion policy in RL.
- Utilizing a learned stimulus frequency policy for adaptive gait generation.
- Employing specialized reward functions for locomotion feature and stimulus correspondence.
- Demonstrating sim-to-real transfer for real-world deployment.
Main Results:
- Achieved robust 3D locomotion and adaptive gait under disturbance for the BITeno robot.
- Demonstrated efficient sim-to-real transfer, enabling real-world performance.
- Exhibited significant disturbance resistance, with recovery times within 1.5-2.0 seconds after a sudden torso velocity change.
Conclusions:
- The proposed RL framework with dynamic stimulus signals effectively enhances bipedal robot adaptability and robustness.
- The method allows for dynamic, model-free gait generation suitable for real-world applications.
- Successful sim-to-real transfer highlights the practical viability of the approach for robust robotic locomotion.
