Event
Research on Reinforcement Learning Control for MARL-Exo Hip Exoskeleton
1 reports · 1 sources
Yifei Yuan and other authors propose the MARL-Exo multi-agent reinforcement learning framework, which trains a full-body musculoskeletal model and bilateral hip exoskeleton policies jointly in simulation in two stages; the exoskeleton observes only delayed hip kinematics and historical commands. In simulation, the metabolic policy reduced modeled muscle metabolic power by 13.8% compared with walking without the device, and by 8.8% after adding mechanical power shaping. In hardware testing, the metabolic policy was used without retraining on eight healthy adults walking at 1.25 m/s, measured by indirect calorimetry. The authors report that MARL-Exo reduced net metabolic rate by 18.3±6.6% compared with the worn-but-unassisted (Passive) condition, with a decrease in all eight participants; compared with no exoskeleton (No-Exo), the average reduction was 7.4±10.0%, among whom six offset the 13.1% wearing penalty, with the lowest being 24.2% below their own walking without the device.
Report timeline
- MARL-Exo: Multi-Agent Reinforcement Learning Co-Adapts Hip Exoskeleton Control to Reduce Metabolic Energy Cost
The study proposes MARL-Exo, a multi-agent reinforcement learning framework in which a full-body musculoskeletal model and a bilateral hip exoskeleton policy are co-trained in simulation in two stages; the exoskeleton observes only delayed hip kinematics and past commands. In simulation, the metabolic policy reduced modeled muscle metabolic power by 13.8% compared with walking without the device, and by 8.8% after adding mechanical-power shaping.
arXiv cs.RO ·