Robust Humanoid Contact Planning With Learned Zero- and One-Step Capturability Prediction

Abstract

Humanoid robots maintain balance and navigate by controlling the contact wrenches applied to the environment. While it is possible to plan dynamically-feasible motion that applies appropriate wrenches using existing methods, a humanoid may also be affected by external disturbances. Existing systems typically rely on controllers to reactively recover from disturbances. However, such controllers may fail when the robot cannot reach contacts capable of rejecting a given disturbance. In this paper, we propose a search-based footstep planner which aims to maximize the probability of the robot successfully reaching the goal without falling under disturbances. The planner considers not only the poses of the planned contact sequence, but also alternative contacts near the planned contact sequence that can be used to recover from external disturbances. Although this additional consideration significantly increases the computation load, we train neural networks to efficiently predict multi-contact zero-step and one-step capturability, which allows the planner to generate robust contact sequences efficiently. Our results show that our approach generates footstep sequences that are more robust to external disturbances than a conventional footstep planner in four challenging scenarios.

Generating Continuous Motion and Force Plans in Real-Time for Legged Mobile Manipulation

“…Additionally, the robot is also able to increase its effective weight by pushing upward on the railing, allowing the robot to exert larger tangential forces at the feet to accelerate forward without having its feet slip on the surface. No forces are applied in the x direction, as we have specified this as a direction of free motion in the constraint in (14).…”

Generating Continuous Motion and Force Plans in Real-Time for Legged Mobile Manipulation

“…This subsection describes a pair of experiments, both of which are included in the atached video, wherein the manipulator forces at the end-effector need to be utilized to ensure that the robot is able to complete a given task without falling over. This is accomplished using the ZMP stability constraint in (17), in which the force at the manipulator's end-effector is directly tied to the stability of the platform and the friction pyramid constraint in (16), which necessitates dynamic feasibility. We do not compare against the baseline planner for these experiments, as the baseline planner is unable to affect the manipulation forces and thus cannot change them to stabilize the robot.…”

“…In contrast, dynamic approaches solve for both the contact location and force of the end-effector for the purpose of keeping the robot stable while it follows a given base trajectory. Such strategies have been used to construct plans to robustly traverse complex terrain [16], [17] as well as for walking and object manipulation [18]. Similarly, methods for footstep planning on uneven terrain involving mixed-integer convex optimization [19] have been adapted to include hand contacts [20].…”

Reachability Map for Diverse Balancing Strategies and Energy Efficient Stepping of Humanoids

“…Previous work produced maps with constraints such as: kinematic reachability [12,13], feasible transitions motions [14,15] and obstacle avoidance [8,16]. Reachability maps can then be used for faster locomotion planning [17,18], complex endpose planning [19], enabling dynamic transitions [20] and for learning [21]. In this paper, we study reachability maps that encode energy cost for reaching step positions across the whole state space and build simple heuristics to capture a diverse range of efficient stepping motions that can be used to plan motions online with little computational cost.…”