Humanoid Loco-manipulation Planning based on Graph Search and Reachability Maps

TL;DR

The paper tackles autonomous loco-manipulation planning for humanoid robots moving large objects in cluttered environments. It introduces a three-process framework (OP-planning, FR-planning, WBM-planning) that combines RRT*-based object path planning with a graph-search-based FR-planning driven by a novel transition model that relabels reachability maps as the object moves. Key contributions include automatic generation of rolling motions with regrasping via reachability-map switching, and a unified planning approach that handles obstacle avoidance, regrasping, and multi-contact dynamics, demonstrated on bobbin rolling, door opening, and cart pushing tasks with dynamic feasibility validated by ZMP analysis. The approach yields fast initial solutions and scalable final plans within seconds, enabling practical real-time-like loco-manipulation planning for humanoids in industrial-like settings. Overall, this work advances humanoid loco-manipulation by integrating object motion, grasp changes, and environment constraints within a single, efficient planning framework, with potential for real-time replanning and extension to more complex interactions.

Abstract

In this letter, we propose an efficient and highly versatile loco-manipulation planning for humanoid robots. Loco-manipulation planning is a key technological brick enabling humanoid robots to autonomously perform object transportation by manipulating them. We formulate planning of the alternation and sequencing of footsteps and grasps as a graph search problem with a new transition model that allows for a flexible representation of loco-manipulation. Our transition model is quickly evaluated by relocating and switching the reachability maps depending on the motion of both the robot and object. We evaluate our approach by applying it to loco-manipulation use-cases, such as a bobbin rolling operation with regrasping, where the motion is automatically planned by our framework.

Figures (15)

• Figure 1: Humanoid loco-manipulation motions.
• Figure 2: Overview of the planning framework. The loco-manipulation planning framework consists of three processes: object path planning (OP-planning), footstep and regrasping planning (FR-planning), and whole-body motion planning (WBM-planning).
• Figure 3: An example of models and results in OP-planning. The shape of the robot is approximated by a bounding box. Five candidates for the robot (in its bounding box) pose relative to the object are shown as an example. The robot pose is discretized and represented as the index of the candidates.
• Figure 4: Problem settings of FR-planning. A robot is moving an object from the start position to the goal position. Green rectangle markers represent the footstep action set for the right foot; a blue line represents the object path; and markers with yellow to red gradation represent the reachability map.
• Figure 5: Transitions in state successors of FR-planning. Orange markers represent the left foot; green markers represent the right foot; and blue markers represent the object. $\mathit{idx}^{[k]}$ is an abbreviation for $\mathit{idx}(\bm{c}_{\mathit{obj}}^{[k]})$.