Footstep planner lets a simulated Digit humanoid avoid obstacles without losing balance

Researchers led by Athanasios Kyriakopoulos have presented a footstep planner for humanoid robots that combines obstacle avoidance with capturability constraints, in a paper accepted for the 2026 IEEE-RAS Humanoids conference.

The sequential model predictive control method builds safe corridors from intersecting rectangles and guarantees recursive feasibility for its planning layer. When an obstacle such as a door that was assumed open turns out to be closed, the planner switches to walking in place.

Its performance depends little on the prediction horizon, which keeps computing costs low. The authors tested it in high-fidelity simulations with the Digit humanoid.