ImpedanceMimic learns per-joint stiffness and damping for force-accurate humanoid motion imitation
carlosdponx · x · 2026-11-18
A Harvard/Kempner Institute team introduces ImpedanceMimic, which treats per-joint stiffness and damping as a learnable action space for force-accurate motion imitation across 129 walking, running and jumping clips. Fixed PD gains can track motion yet miss contact forces; learned impedance keeps force fidelity as motions get agile. Commenter carlosdponx notes learned PD control hasn't been tried in published research for a while, and cheaper training makes it worth revisiting.
More from Embodied
- Argon open-sources Alloy, a multi-material physics engine for robot learning on one GPU — ycombinator · 2026-11-18
- California DMV Orders Waymo Off State Freeways After Traffic Incidents — Polymarket · 2026-11-18
- StackChan desk robot gets a major vision boost from a 5MP ESP32-S3 camera — alexcovo_eth · 2026-11-18
- OpenAI reportedly moves robotics researchers into general AI teams as models learn to control robots — Dr_Singularity · 2026-11-18
- $160 ChatGPT-powered kids' toy Contixo Pixie-Paw pulled after telling stories of sexual violence — dailymail · 2026-11-18
- How one developer built a robot companion in 24 hours with Astra — gabrielchua · 2026-11-18