Robots can work together to cooperatively execute tasks much faster than a single robot. In the scenario proposed by Jones et al. fire trucks are sent out to extinguish fires caused by a large-scale disaster. Because of the disaster, roads are blocked by debris that can only be cleared by bulldozer robots. Coordination in this scenario amounts to figuring out which routes the fire trucks should take to extinguish which fires and how bulldozers should be used to clear the way. Good coordination leads to a maximum number of fires being extinguished as fast as possible.
Allocating the tasks to the different agents (fire trucks and bulldozers) over time is challenging because of the explosion in possible combinations of agents, tasks and routes. To address this challenge, Jones et al. propose two approaches. In the first, agents bid on groups of tasks to be accomplished over time and auctions are then held to distribute the tasks. The second approach searches over all possible solutions by using a genetic algorithm.
Experiments in simulation show that the genetic algorithm, if given enough time, results in better system performance than auction-based systems that tend to result in local minima. Higher performance however comes at the price of orders of magnitude increase in processing. Because both approaches are able to achieve good solutions, the tradeoff between performance and execution time will need to be considered on a case by case basis.
Two examples of auction-based approaches are shown below. On the left side, only a single fire is assigned per fire truck at a time, while the right side approach allows several fires to be assigned at a time. Result show that assigning a set of tasks to accomplish over a period of time leads to better performance (green bar) than assigning a single task at a time (time-extended coordination).