Heuristic Planner for Communication-Constrained Multi-Agent Multi-Goal Path Planning

TL;DR

The paper tackles the problem of coordinating multiple agents to visit an ordered sequence of goals under a communication constraint, framing it as a CC-PP problem on a directed weighted graph. It introduces a two-stage heuristic graph-search planner: Stage 1 computes epoch-specific heuristics by designating a leader to reach the current goal while followers determine feasible end locations that maintain communication; Stage 2 performs a greedy best-first search using a cost-based heuristic that aggregates per-agent costs raised to a power $\alpha$. The approach demonstrates the ability to handle up to about ten agents on several maps, with makespan decreasing as agent count and communication range increase, though the first stage dominates computation and performance can degrade in challenging maps or loose constraints. This central, stage-based planning framework offers a path toward practical swarm coordination under communication constraints and could integrate with roadmap-based representations and alternative communication metrics in future work.

Abstract

In robotics, coordinating a group of robots is an essential task. This work presents the communication-constrained multi-agent multi-goal path planning problem and proposes a graph-search based algorithm to address this task. Given a fleet of robots, an environment represented by a weighted graph, and a sequence of goals, the aim is to visit all the goals without breaking the communication constraints between the agents, minimizing the completion time. The resulting paths produced by our approach show how the agents can coordinate their individual paths, not only with respect to the next goal but also with respect to all future goals, all the time keeping the communication within the fleet intact.

Figures (6)

• Figure 1: Example behavior of the agents. ●: starting location, ♦: locations at the time the first goal is reached, ✖: final positions. While the red agent visits the first goal, the other two agents position themselves favorably with respect to the second goal, maintaining the communication distance as they do so.
• Figure 2: Example of Stage 1 computation for a simple map. Darkest color represents obstacles. Blue dots represent goals. The shaded areas show the nodes explored by the agent (lighter color corresponds to nodes opened later in time). Orange shading shows result nodes, red shading (crosses) represents potential starting nodes (sometimes not visible under the result node marks).
• Figure 3: Maps: Paris, (left to right) Blobs, Maze. Red and blue marks show goals and starting position (agents start close together around this point). Traces of the agents shown in colors. Potential communication links in white, communication limit in red in the top left corner. Each white cluster represents a different time snapshot, and represents the full communication graph.
• Figure 4: The makespan with respect to the number of agents and the communication range for the map Blobs (column with 1 agent is just a single agent shortest tour visiting all the goals for comparison). Missing value is the instance where none of the experiments finished successfully.
• Figure 5: Time requirements of the individual computational steps, normalized. Showing result on the Blobs map, average values over all tested communication limits.