Conflict-Based Search for Explainable Multi-Agent Path Finding (2202.09930v2)

Abstract: In the Multi-Agent Path Finding (MAPF) problem, the goal is to find non-colliding paths for agents in an environment, such that each agent reaches its goal from its initial location. In safety-critical applications, a human supervisor may want to verify that the plan is indeed collision-free. To this end, a recent work introduces a notion of explainability for MAPF based on a visualization of the plan as a short sequence of images representing time segments, where in each time segment the trajectories of the agents are disjoint. Then, the explainable MAPF problem asks for a set of non-colliding paths that admits a short-enough explanation. Explainable MAPF adds a new difficulty to MAPF, in that it is NP-hard with respect to the size of the environment, and not just the number of agents. Thus, traditional MAPF algorithms are not equipped to directly handle explainable-MAPF. In this work, we adapt Conflict Based Search (CBS), a well-studied algorithm for MAPF, to handle explainable MAPF. We show how to add explainability constraints on top of the standard CBS tree and its underlying A* search. We examine the usefulness of this approach and, in particular, the tradeoff between planning time and explainability.

• The paper introduces Explanation-Guided CBS (XG-CBS), adapting Conflict-Based Search to integrate explainability constraints for generating explainable Multi-Agent Path Finding plans.
• It proposes three low-level search algorithms, XG-A*, WXG-A*, and SR-A*, each balancing computational complexity with plan index optimization differently.
• Empirical validation demonstrates that these methods improve scalability and segment reduction, enhancing trust in AI systems for safety-critical applications.

A Formal Overview of Conflict-Based Search for Explainable Multi-Agent Path Finding

The paper of Multi-Agent Path Finding (MAPF) is paramount for many AI applications where multiple agents navigate an environment simultaneously. The fundamental goal is to find paths that are devoid of collisions to ensure that each agent reaches its target destination. However, this complexity is magnified in safety-critical domains, where the veracity of such plans must often be scrutinized by human supervisors. The paper "Conflict-Based Search for Explainable Multi-Agent Path Finding" introduces a nexus between explainability and MAPF by proposing a scheme that binds the existing Conflict-Based Search (CBS) algorithm with explainable AI paradigms.

Overview

The research addresses the inherently complex nature of Explainable MAPF by augmenting the CBS algorithm to accommodate constraints ensuring that the resultant paths can be readily segmented into a minimal number of disjoint segments. This segmentation plays a crucial role in visualizing the safety of the plan, providing a human-understandable verification mechanism. The central challenge addressed is that of NP-hardness in respect to environmental size; traditional MAPF approaches inadequately address this dimension of explainability.

Key Contributions

The authors make several noteworthy contributions:

1. Adaptation of CBS: The paper expands the classical CBS algorithm by integrating explainability constraints on top of the CBS search tree, effectively transforming it to handle Explainable MAPF problems efficiently.
2. Explanation-Guided CBS (XG-CBS): This is the primary contribution, where the authors propose a new version of CBS that integrates segmentation conflicts. These conflicts arise when a plan results in more segments than the specified bound and are resolved by placing additional constraints on the CBS nodes.
3. Low-Level Search Algorithms: Three low-level search algorithms are introduced and juxtaposed - namely, XG-$A^*$, WXG-$A^*$, and SR-$A^*$. Each of these algorithms attempts to navigate the trade-offs between computational complexity and plan index optimization distinctively. XG-$A^*$ focuses on minimizing segments through history tracking, WXG-$A^*$ attempts a balanced approach, and SR-$A^*$ forsakes completeness for performance gains.
4. Empirical Validation: A series of rigorous benchmarks and experiments elucidate the comparative advantages and limitations of their algorithmic solutions, highlighting the improved scalabilty and segment reduction of the Explainable MAPF solutions over conventional methodologies.

Implications and Future Directions

By establishing a framework that rigorously combines explainable AI with path planning, this research opens up multiple avenues for both theoretical exploration and practical applications. The presented methods significantly enhance trust between human supervisors and AI systems, particularly in regulated environments where operational transparency and safety assurances are prioritized.

Future explorations can include the extension of this method to more sophisticated MAPF variants and integrating further AI explainability techniques. Applying such methods in domains beyond traditional MAPF applications, like tethered robot navigation or layered circuitry design, could provide practical benefits and cost reductions. Furthermore, evolving beyond CBS, exploring other efficient search-based MAPF solvers with an eye for explainability, could yield further advancements in this space.