Massively Parallel Sim-to-Real for Humanoid Locomotion
日本語版は未提供のため、英語原文で表示しています。
30秒で読む
RL policies trained in GPU simulation transfer zero-shot to a commodity humanoid on rough terrain.
研究課題
Does domain randomisation scale to full humanoids?
問題
Humanoid balance is brittle under sim/real gaps.
従来の手法
Model-based controllers with hand-tuned gains.
新しい手法
Terrain curricula plus actuator-network modelling.
結果
Stairs, slopes and pushes on a Unitree platform.
限界
Locomotion only; no manipulation.
産業への影響
Lowers the barrier for low-cost humanoids to walk reliably, compressing hardware differentiation.