Multi-contact Planning

Planning which body part — foot, hand, knee — contacts the environment where and in what order, to make use of that contact for leverage.

Multi-contact planning is about deciding, for a legged or humanoid robot, which body part touches the environment, when, and where, in a way that's mechanically feasible for the whole motion. Ordinary bipedal walking alternates just two feet; climbing a steep slope, squeezing through a tight gap, getting over an obstacle, or catching a fall may call for a hand against a wall or a knee on the ground too, with neither the number nor the order of contact points fixed in advance. The difficulty is that it mixes discrete decisions (which limb, what order, which surface to land on) with continuous ones (the body trajectory, keeping contact forces inside their friction cones), making it a mixed discrete-continuous optimization problem. Traditional approaches split it into layers: plan the contact sequence and positions first, then generate the motion with centroidal dynamics or whole-body trajectory optimization; recent work uses Monte Carlo tree search or a learned value function to combine the two layers, run in a receding-horizon, online-replanning fashion. It is in the same family as footstep planning, but without being restricted to a periodic gait.

ExampleA Talos humanoid walking up a slope too steep to balance statically on: Wang et al.'s 2023 method uses a learned value function to approximate long-term consequences, planning the next contact and body motion online in a receding horizon.

Also called

Contact Planning

Related

Footstep Planning · Gait Planning · Centroidal Dynamics · Friction Cone · Contact-Implicit Trajectory Optimization · Monte Carlo Tree Search

Sources

Simultaneous Contact Sequence and Patch Planning for Dynamic Locomotion (arXiv 2508.12928)
Online Multi-Contact Receding Horizon Planning via Value Function Approximation (arXiv 2306.04732)