Conflict-Based Search and Prioritized Planning for Multi-Agent Path Finding Among Movable Obstacles

TL;DR

This work defines Multi-Agent Path Finding Among Movable Obstacles (M-PAMO), where agents must navigate a grid containing static and movable obstacles that can be pushed. It adapts Conflict-Based Search (CBS) and Prioritized Planning (PP) by proposing CBS-MOH, CBS-MOL, and PP-PAMO*, with ST-PAMO* as a time-augmented PAMO* extension for the low-level; these methods are evaluated on maps with up to 20 agents and hundreds of movable obstacles. The results show that CBS-MOH and CBS-MOL are not guaranteed to be complete or optimal for M-PAMO and reveal trade-offs between high-level conflict reasoning and low-level box awareness. The study highlights wakeful performance differences between CBS-based and PP-based strategies in tightly coupled scenarios, pointing to future work on richer information sharing about moved boxes and adaptive prioritization to improve scalability and solution quality.

Abstract

This paper investigates Multi-Agent Path Finding Among Movable Obstacles (M-PAMO), which seeks collision-free paths for multiple agents from their start to goal locations among static and movable obstacles. M-PAMO arises in logistics and warehouses where mobile robots are among unexpected movable objects. Although Multi-Agent Path Finding (MAPF) and single-agent Path planning Among Movable Obstacles (PAMO) were both studied, M-PAMO remains under-explored. Movable obstacles lead to new fundamental challenges as the state space, which includes both agents and movable obstacles, grows exponentially with respect to the number of agents and movable obstacles. In particular, movable obstacles often closely couple agents together spatially and temporally. This paper makes a first attempt to adapt and fuse the popular Conflict-Based Search (CBS) and Prioritized Planning (PP) for MAPF, and a recent single-agent PAMO planner called PAMO*, together to address M-PAMO. We compare their performance with up to 20 agents and hundreds of movable obstacles, and show the pros and cons of these approaches.

Figures (8)

• Figure 1: An example of M-PAMO. (a) shows the problem instance. (b) shows that if agent 1 follows its shortest path and reaches its goal at time $t=3$, there is no way for agent 2 to reach its goal due to the movable obstacle $B1$, since it cannot push $B1$ onto agent 1 at time $t=4$. (c) shows the solution returned by our CBS-MOH and CBS-MOL, where agent 1 has to wait for 2 steps at first and then pushes $B1$ to the right so that both agents can then reach their goals.
• Figure 2: Different types of conflicts in M-PAMO.
• Figure 3: An example where adding constraints onto the boxes fails. (a) shows the instance. (b) shows a conflict at t=2. (c) and (d) show the low-level planning: CBS-MOH ignores the box, while CBS-MOL still regards the box at its original position. (e) shows the constraint tree where the added constraint fails to resolve the conflicts and the search never terminates. More details can be found in Sec. \ref{['M_PAMO:sec:discuss']}.
• Figure 4: M-PAMO examples. The first column shows 4 different instances. The 2nd and 3rd column show the planned results of CBS-MOH and CBS-MOL respectively. Row (a) shows a case where CBS-MOH fails while CBS-MOL succeeds. Row (b) shows a case where CBS-MOH succeeds while CBS-MOL fails. Row (c) shows a case where both planners succeed but find different paths, and CBS-MOL fails to find an optimal solution. Row (d) shows a case where CBS-MOH fails to find an optimal solution.
• Figure 5: Test instance examples. (a) shows an $8\times8$ empty grid map without static obstacles and with 30% box density, and (b) shows a $16\times16$ random grid map with 10% static obstacles and 30% box density.