사족보행 로봇 연구 기록
← 피드로 돌아가기
V1 · 소프트웨어 · Oct 15, 2026

Analysis: walking speed on flat ground

The flat policy, the stair policy and the three multi-terrain policies walked the same 50 flat-ground trials at 0.3, 0.5 and 0.75 m/s, to see what training on terrain costs in speed.

  • At a 0.5 m/s command: flat policy 0.44 m/s, multi-terrain policies started from the flat policy 0.43 and 0.42, the one started from the stair policy 0.40, stair policy 0.39
  • At 0.75 m/s every terrain-trained policy is slower than the flat policy: 0.51–0.56 against 0.60 m/s
  • The multi-terrain policies sway least: body roll rate 0.41–0.60 rad/s against 0.76 for the flat policy and 0.80 for the stair policy
  • The stair policy still climbs most: 4 cm stairs 82 of 100, against 77 and 63 for the multi-terrain policies started from the stair and the flat policy, and 0 for the flat policy
  • The speed is lost in stair-only training, not to the body-height reward (stair policies trained without it walk 0.39–0.41 m/s) and not to short treads (treads up to 300 mm: 0.40 m/s)