Design and implementation of consensus in multi-agent systems with input and communication time-delays
No Thumbnail Available
Achieving consensus is a major problem in multi-agent systems(MASs). Analysis, design and implementation of consensus in MASs is the main focus of this thesis. Motivated by recent surge in distributed control applications with networked mobile robots and sensor networks which are MASs, consensus problems in autonomous mobile robots and a sensor network are researched. In particular, faults in communication like complete loss and time-delay effects on consensus are studied. Complete loss of communication will prevent the multi-agent system in reaching consensus. Time-delay within certain limit deteriorate the performance of system by increasing the convergence time. Beyond the limit, the system never reaches consensus.Two novel algorithms, namely “back-tracking” and “history following”, to reach consensus in case of communication loss for a multi-agent system(MAS) with switching topologies are proposed for mobile agents. A strongly connected topology is considered to make decision on usage of “back-tracking” algorithm whenever an agent loses communication. It ensures the mobile multiagent consensus with intermittent communication loss by changing path of the agents to maintain a strongly connected topology. In the case of “history following” algorithm, the agents store past data and continue with most recent target position even after the loss of communication. Since the communication loss that we have considered is intermittent at few places, the agents regain communication after some displacement. Then the data is updated with new calculation of target position and stored for further usage that might arise. A simple obstacle avoidance algorithm also used while simulating a network of six mobile agents with intermittent communication loss and static obstacles. The pros and cons of both algorithms are discussed. Hardware implementation is performed for a three robot system without obstacles along with corresponding simulation to demonstrate the effectiveness of the algorithms.
Supervisors: Indrani Kar and Somanath Majhi
ELECTRONICS AND ELECTRICAL ENGINEERING