Massively Parallel Sim-to-Real for Humanoid Locomotion
暂无中文版，显示英文原文。
30 秒速读
RL policies trained in GPU simulation transfer zero-shot to a commodity humanoid on rough terrain.
研究问题
Does domain randomisation scale to full humanoids?
问题
Humanoid balance is brittle under sim/real gaps.
既有方法
Model-based controllers with hand-tuned gains.
新方法
Terrain curricula plus actuator-network modelling.
结果
Stairs, slopes and pushes on a Unitree platform.
局限
Locomotion only; no manipulation.
产业影响
Lowers the barrier for low-cost humanoids to walk reliably, compressing hardware differentiation.