Massively Parallel Sim-to-Real for Humanoid Locomotion
아직 한국어판이 없어 영어 원문으로 표시.
30초 요약
RL policies trained in GPU simulation transfer zero-shot to a commodity humanoid on rough terrain.
연구 질문
Does domain randomisation scale to full humanoids?
문제
Humanoid balance is brittle under sim/real gaps.
기존 접근
Model-based controllers with hand-tuned gains.
새 접근
Terrain curricula plus actuator-network modelling.
결과
Stairs, slopes and pushes on a Unitree platform.
한계
Locomotion only; no manipulation.
업계 영향
Lowers the barrier for low-cost humanoids to walk reliably, compressing hardware differentiation.