CE-MRS: Contrastive Explanations for Multi-Robot Systems

TL;DR

CE-MRS tackles explainability in heterogeneous multi-robot systems where the solution is a triple $\mathcal{S}=\langle \mathcal{A}, \sigma, \mathcal{M} \rangle$. It introduces contrastive explanations $\mathcal{E_S}$ that compare $\mathcal{S}$ to a foil $\mathcal{S'}=\langle \mathcal{A'}, \sigma', \mathcal{M'}\rangle$, enabling feasibility checks and performance-based contrasts. It formalizes the framework, selects information from task allocation, scheduling, and motion planning, and validates with a 22-participant study in a search-and-rescue domain, showing improved error identification and remediation when using CE-MRS. The findings indicate that cross-subproblem explanations yield more informative feedback and enhance operator performance in complex multi-robot tasks.

Abstract

As the complexity of multi-robot systems grows to incorporate a greater number of robots, more complex tasks, and longer time horizons, the solutions to such problems often become too complex to be fully intelligible to human users. In this work, we introduce an approach for generating natural language explanations that justify the validity of the system's solution to the user, or else aid the user in correcting any errors that led to a suboptimal system solution. Toward this goal, we first contribute a generalizable formalism of contrastive explanations for multi-robot systems, and then introduce a holistic approach to generating contrastive explanations for multi-robot scenarios that selectively incorporates data from multi-robot task allocation, scheduling, and motion-planning to explain system behavior. Through user studies with human operators we demonstrate that our integrated contrastive explanation approach leads to significant improvements in user ability to identify and solve system errors, leading to significant improvements in overall multi-robot team performance.

Figures (4)

• Figure 1: CE-MRS Framework Diagram
• Figure 2: Our user interface, in which the System Solution Panel (left) visualizes the system solution $\mathcal{S}$ for domain definition $\mathcal{D}$ (a). The User Sandbox Panel (right) visualizes a user's foil solution $\mathcal{S'}$ (b) for their foil allocation $\mathcal{A'}$ (c).
• Figure 3: RSE%, RTE%, and TRE% metrics per study condition, in which a lower value is better. Statistical significance is reported as: * p$<$0.01, ** p$<$0.001
• Figure 4: Efficiency of users' error corrections of $\mathcal{D}$. Points further left are scenarios with more errors corrected; points closer to the line represent more efficient corrections.