BROWSE
Detail
Conference Paper
뱀 로봇의 강화학습 효율 향상을 위한 토크 기반 게이트 분해 기법 및 탐색 공간 축소 효과 분석
Cited - time in 
Share 
Share
- Authors
- 송봉섭, 윤동원
- Issue Date
- 2026-02
- Citation
- 한국로봇학회 종합 학술 대회 2026
- Publisher
- 한국로봇학회
- Language
- Korean
- Type
- Conference Paper
- Abstract
- This study proposes a Torque-based Gait Decomposition (GD) method to improve the reinforcement learning (RL) efficiency of snake robots with high degrees of freedom. By decomposing the conventional Serpenoid curve into a Motion Matrix representing torque direction and a neural network approximating torque magnitude, the proposed method significantly reduces the exploration space of the RL agent. We theoretically analyze the reduction ratio of the action space and experimentally verify that the GD-based approach achieves faster convergence and higher stability compared to baseline PPO and SAC algorithms in complex gait learning tasks.
- Keyword
- Bio-Inspired Robotics, Reinforcement Learning, Snake Robot
- KSP Keywords
- Action space, Based Approach, Bio-Inspired Robotics, Degrees of freedom(DOF), Faster convergence, Gait learning, Reinforcement learning(RL), Snake robot, Torque-Based, high degrees of freedom, neural network(NN)