ACLM: ADMM-Based Distributed Model Predictive Control for Collaborative Loco-Manipulation

TL;DR

This study proposes an Alternating Direction Method of Multipliers (ADMM)-based distributed model predictive control framework for collaborative loco-manipulation with a team of quadruped robots with manipulators that achieves fast convergence, requiring only a few ADMM iterations per planning cycle.

Abstract

Collaborative transportation of heavy payloads via loco-manipulation is a challenging yet essential capability for legged robots operating in complex, unstructured environments. Centralized planning methods, e.g., holistic trajectory optimization, capture dynamic coupling among robots and payloads but scale poorly with system size, limiting real-time applicability. In contrast, hierarchical and fully decentralized approaches often neglect force and dynamic interactions, leading to conservative behavior. This study proposes an Alternating Direction Method of Multipliers (ADMM)-based distributed model predictive control framework for collaborative loco-manipulation with a team of quadruped robots with manipulators. By exploiting the payload-induced coupling structure, the global optimal control problem is decomposed into parallel individual-robot-level subproblems with consensus constraints. The distributed planner operates in a receding-horizon fashion and achieves fast convergence, requiring only a few ADMM iterations per planning cycle. A wrench-aware whole-body controller executes the planned trajectories, tracking both motion and interaction wrenches. Extensive simulations with up to four robots demonstrate scalability, real-time performance, and robustness to model uncertainty.

Figures (7)

• Figure 1: ADMM-based distributed MPC framework exploiting the star-shaped coupling induced by the shared payload. The robot subproblems are solved in parallel with consensus on interaction wrenches, and the resulting trajectories are executed by local WBC.
• Figure 2: Illustration of diverse terrain navigation scenarios used to evaluate collaborative loco-manipulation. A.1--A.2: Two robots carry a cargo across stepped terrain of uniform surface height, shown in rendered and RViz views. B.1--B.2: Two robots carry a cargo on a sloped surface, shown in rendered and RViz views. C: Two robots carry a cargo through a 90-degree turn in a narrow passage. D: Two robots carry a cargo following a circular path on an elevated annular platform. E: Four robots carry a folding stretcher across stepped terrain of uniform surface height. F: Three robots carry a wooden table on a sloped surface. G: Two robots carry a cargo performing obstacle avoidance on flat terrain.
• Figure 3: CPU time scalability comparison between distributed and centralized MPC for 2–4 robots in flat and Gap-Slope terrain scenarios.
• Figure 4: Residual convergence (top-left, top-right, and bottom-left) and computation time (bottom-right) for different ADMM-SQP iterations.
• Figure 5: Obstacle avoidance and pose tracking performance.

Theorems & Definitions (1)

• Remark 1