Event
A Perception Method for Humanoid Robots to Anticipate Footfall Locations
1 reports · 1 sources
Matthias Heyrman et al. released a paper on arXiv proposing a lightweight footfall prediction model (FPM): it predicts future footfall locations using only proprioception and velocity commands, samples elevation maps only around these footfall locations, and lets perception anticipate foot placement ahead of the policy. The paper claims that this strategy transfers zero-shot to the Unitree G1 humanoid robot, traversing terrains beyond the training distribution such as indoor and outdoor obstacle crossing, stairs, trenches, and step height differences; with this preprocessing, the training iterations required to reach the same terrain difficulty are about one-third of those needed when using the full elevation map. The prediction model is trained together with the locomotion policy and is periodically retrained with new policy rollouts as the robot's gait matures.
Report timeline
- Watch Your Step: A perception method that anticipates footholds improves humanoid terrain locomotion
The arXiv paper Watch Your Step proposes a lightweight foothold prediction model (FPM) that predicts future footholds from proprioception and velocity commands alone, and samples the elevation map only around those footholds, letting perception anticipate foot placement ahead of the policy.
arXiv cs.RO ·