Optimal Distributed Multi-Robot Communication-Aware Trajectory Planning using Alternating Direction Method of Multipliers

TL;DR

The paper tackles maintaining connectivity in distributed multi-robot trajectory planning under communication constraints by formulating a non-separable connectivity constraint based on the Fiedler value. It introduces an economic interpretation that separably distributes a communication budget and enables trading among neighboring robots, solved with a dual-ascent consensus ADMM approach. A local adjacency estimation and M-step prediction framework allows robots to coordinate without a central coordinator, and convergence is achieved via a wavefront-like consensus protocol. Simulation in an inspection task demonstrates that budget trading preserves connectivity while reducing collective cost, offering a scalable approach with potential applicability to other shared-resource constraints.

Abstract

This paper presents a distributed, optimal, communication-aware trajectory planning algorithm for multi-robot systems. Building on prior work, it addresses the multi-robot communication-aware trajectory planning problem using a general optimisation framework that imposes linear constraints on changes in robot positions to ensure communication performance and collision avoidance. In this paper, the optimisation problem is solved distributively by separating the communication performance constraint through an economic approach. Here, the current communication budget is distributed equally among the robots, and the robots are allowed to trade parts of their budgets with each other. The separated optimisation problem is then solved using the consensus alternating direction method of multipliers. The method was verified through simulation in an inspection task problem.

Figures (8)

• Figure 1: An example of an ad-hoc network where five drones communicate wirelessly with each other and a ground station antenna. Three drones outside the antenna's range (blue dashed line) rely on each other to relay their information Mikkelsen2024OptimalValue.
• Figure 2: Link quality with increasing values of $\alpha$, in order of red, green, and blue Mikkelsen2024OptimalValue.
• Figure 3: D-MR-CaTP algorithm structure.
• Figure 4: Snapshots of simulation at four different iterations of D-MR-CaTP algorithm with 5 inspection points and $N=10$ robots. The circles indicate inspection robots while the triangles indicate support robots. The POIs are indicated by the black circles with the numbering of the POI being adjacent.
• Figure 5: Snapshots of simulation at four different iterations of D-MR-CaTP algorithm with 5 inspection points and $N=10$ robots. The circles indicate inspection robots while the triangles indicate support robots. The POIs are indicated by the black circles with the numbering of the POI being adjacent.