Collaborative decision making in uncertain environments

• Main Author: Baxter, Joseph L.
• Published: University of Nottingham 2009
• Subjects: 629.8; TJ Mechanical engineering and machinery
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.514747

Two major issues in the design of multi-robot systems are those of communication and co-ordination. Communication ithin real world environments cannot always be guaranteed. A multi-robot system must, therefore, be able to continue with its task in the absence of communication between team members. Co-ordination of multiple robots to perform a specific task involves team members being able to make decisions as a single entity and as a member of a team. The co-ordination needs to be robust enough to handle failures within the system and unknown phenomena within the environment. In this thesis, the problems of communication and co-ordination are discussed and a new type of multi-robot system is introduced in an effort to solve the inherent difficulties within communication and co-ordination of multi-robot systems. The co-ordination and communication strategy is based upon the concept of sharing potential field information within dynamic local groups. Each member of the multi-robot system creates their own potential field based upon individual sensor readings. Team members that are dynamically assigned to local groups share their individual potential fields, in order to create a combined potential field which reduces the effect of sensor noise. It is because of this, that team members are able to make better decisions. A number of experiments, both in simulation and in laboratory environments, are presented. These experiments compare the performance of the system against a nonsharing control and a hybrid system made up of a global path planner and a reactive motor controller. It is demonstrated that the new system significantly outperforms these other methods in a search type problem. From this, it is concluded that the novel system proposed in this thesis successfully tackled the search problem, and that it should also be possible for the system to be applied to a number of other common multi-robot problems.