Applying the Gale-Shapley Stable Matching Algorithm to Peer Human-Robot Task Allocation

When human and robotic agents work together, the challenge in assigning tasks lies in exploiting human strengths, such as expertise and intuition, while still managing the heterogeneous agent team in a near-optimal way. An extension to the Gale-Shapley stable matching algorithm that combines a sequential greedy approach is proposed to apply to task allocation missions. Conventional task features are modeled in the form of task preferences; agent inputs are modeled in the form of agent preferences. The algorithm is applied to a bomb defusal scenario, where bomb location is known but time for each agent to defuse each bomb is supplied through agent preferences. Simulation results are presented, and the sequential greedy Gale-Shapley algorithm is compared to a corresponding sequential single-item auction algorithm under three evaluation criteria — mission completion time, agent-task pair regret, and evenness of task distribution among agents.