TactileStep Makes Humanoid Landings Softer
Tsinghua’s TactileStep adds sole-pressure feedback to Unitree G1 locomotion policies, using contact force, center of pressure, and loaded-area signals to regulate touchdown and stance. Tests across varied terrain report up to 48.8% lower touchdown force and 23.8% greater stance contact area. [Official project page](https://tactilestep.github.io/) · [Paper](https://arxiv.org/abs/2609.28959)
TactileStep makes a strong case that vision alone is insufficient for reliable humanoid locomotion: seeing a foothold is different from knowing how the foot actually loaded it. The results are promising, but the system remains a research prototype with hardware, sensing-bandwidth, and energy tradeoffs.
- –Compact tactile features let the policy close the loop on real contact without processing the full pressure map.
- –Reported gains include up to 30.1 dB lower impact noise and broader, more centered support across tested terrains.
- –The approach was validated on a real Unitree G1, strengthening its sim-to-real credibility.
- –Simulated policies generally used more energy and moved slightly slower than the vision-based baseline.
- –Code, weights, long-term sensor testing, and deformable-terrain evaluations were not yet available. [Independent technical review](https://roboskin.ai/news/tactilestep-sole-pressure-humanoid-locomotion)
DISCOVERED
1h ago
2026-10-04
PUBLISHED
1h ago
2026-10-04
RELEVANCE
AUTHOR
AI Search