Task Elimination: Faster Coalition Formation for Overtasked Collectives

Robotic collectives for disaster response require coalition formation algorithms to generate high-quality robot team to task assignments in at most near-real time (i.e, approximate to 5 minutes). Hedonic game-based algorithms have had some success in meeting these criteria, overcoming the long worst-case runtimes of traditional auction methods. However, hedonic games are not well-studied for overtasked collectives (i.e., collectives assigned more tasks than they can perform). A key risk is that hedonic games may produce low-quality solutions by assigning insufficient robots to several tasks, rather than sufficient robots to a task subset. Task elimination (i.e., removing tasks until all remaining tasks' requirements are met) is proposed, incorporating two heuristics: eliminating the minimum value task and using difference rewards. A distributed simulation-based evaluation with 100 robots demonstrates that task elimination produces substantially higher-quality solutions than a straight hedonic game, while maintaining similarly low runtimes (i.e., < 40 seconds). Task elimination is also substantially faster than the auction baseline, with a reasonable solution quality tradeoff.