CTS-CBS: A New Approach for Multi-Agent Collaborative Task Sequencing and Path Finding

TL;DR

CTS-MAPF extends MAPF by requiring coordinated sequencing of intermediate tasks in addition to collision-free paths. The authors introduce CTS-CBS, a two-level, CBS-inspired framework that combines a high-level $K$-best jTSP-based search forest with a low-level constrained path planner for each agent, and prove completeness and $\omega$-bounded suboptimality. A novel method to compute the $K$-best joint task sequences and three MG-MAPF adaptations are presented, with extensive experiments on CTS-MAPF and MG-MAPF datasets showing substantial gains in success rate and runtime while incurring limited loss in solution quality; practical robot tests corroborate real-world applicability. The work enables flexible trade-offs between optimality and efficiency via the suboptimality bound $\omega$, and provides a scalable approach to jointly optimize task sequencing and collision-free motion in multi-agent systems.

Abstract

This paper addresses a generalization problem of Multi-Agent Pathfinding (MAPF), called Collaborative Task Sequencing - Multi-Agent Pathfinding (CTS-MAPF), where agents must plan collision-free paths and visit a series of intermediate task locations in a specific order before reaching their final destinations. To address this problem, we propose a new approach, Collaborative Task Sequencing - Conflict-Based Search (CTS-CBS), which conducts a two-level search. In the high level, it generates a search forest, where each tree corresponds to a joint task sequence derived from the jTSP solution. In the low level, CTS-CBS performs constrained single-agent path planning to generate paths for each agent while adhering to high-level constraints. We also provide heoretical guarantees of its completeness and optimality (or sub-optimality with a bounded parameter). To evaluate the performance of CTS-CBS, we create two datasets, CTS-MAPF and MG-MAPF, and conduct comprehensive experiments. The results show that CTS-CBS adaptations for MG-MAPF outperform baseline algorithms in terms of success rate (up to 20 times larger) and runtime (up to 100 times faster), with less than a 10% sacrifice in solution quality. Furthermore, CTS-CBS offers flexibility by allowing users to adjust the sub-optimality bound omega to balance between solution quality and efficiency. Finally, practical robot tests demonstrate the algorithm's applicability in real-world scenarios.

Figures (8)

• Figure 1: A simple example of CTS-MAPF. There are three agents ($a_i, i=1,2,3$) at their initial locations; their respective goal locations are denoted as (($v_di, i=1,2,3$) and the task locations are denoted as $v_ti$. The colors marked on the task locations indicate the specific agents required to reach there. For example, $v_t1$ requires agents $a_1$ and $a_2$ to reach, while $v_t2$ requires all three agents to reach it. For task locations that require multiple agents, there is no specified order in which the agents must obey. The agents need to plan collision-free paths to visit their respective task locations and ultimately return to their destinations.
• Figure 2: The possible applicable scenarios of CTS-MAPF. (a) for hazardous monitoring and (b) for unmanned market.
• Figure 3: The main process of CTS-CBS. The joint path $\Pi$ of every node $P$ in search tree $CT_j$ follows the same joint task sequence $\mathcal{\Tau}_j$. The node expansion within a single $CT$ is as CBS to resolve conflicts. If there is a need for a new search tree, the new root node is generated to construct it based on the next best joint task sequence obtained by $K$-best joint task sequencing. CTS-CBS operates in a best-first manner, ensuring that the first node containing a conflict-free path $\Pi$ found has the minimum cost $g$.
• Figure 4: The illustration of Algorithm \ref{['algo:kbestjoint']}. In each iteration, the best node from OPENj is popped, followed by adding its extensions into OPENj. By iterating this process, Algorithm \ref{['algo:kbestjoint']} is capable of producing the first, second, ..., $k$-th optimal solutions for joint task sequencing.
• Figure 5: The maps used in experiments. The images also visualize one specific instance on different maps when the number of agents is 5 and the number of tasks is 10. Circles of different colors represent the agents' start locations, while stars denote their goal positions. Red squares represent tasks, with the numbers beside them indicating which agents need to visit them.

Theorems & Definitions (5)

• Definition 1
• Theorem 1
• proof
• Theorem 2
• proof