Next: Mapping and Navigation Up: Control Architectures Previous: Hybrid Architectures
Regarding the use of bidding mechanisms, we have found very few systems making use of it. At CMU, the FIRE project [19] uses a market-oriented approach to model the co-operation of a team of robots. In this approach, instead of using the bidding mechanism to coordinate the agents of a single robot, bidding is used to coordinate a team of robots that have to accomplish several tasks. The rationale is that with this mechanism, each task is performed by the best suited robot for the task, thus achieving a better global performance.
Sun and Sessions [63] have also proposed an approach for developing a multi-agent reinforcement learning system that uses a bidding mechanism to learn complex tasks. The bidding is used to decide which agent gets the control of the learning process. The agents bid according to the expected reward that would receive if they were given the control. Thus, although they are competing for the control, they also cooperate, since they seek to maximize the overall system reward.