Towards collective manipulation without inter-agent communication

This article details the formulation of a decentralized controller for collective manipulation that does not require any communication between agents involved in the task. First, a centralized controller based on the complete system Jacobian is discussed as a benchmark. Then, the centralized controller is reformulated to obtain the algorithm for the proposed decentralized control approach. Both, centralized and decentralized controllers utilize Moore-Penrose pseudoinverse to distribute a control action through the agents of the group. The convergence and stability of both controllers are discussed in detail. Moreover, robustness and effectiveness of the proposed controllers are investigated through simulating numerous scenarios, formations and populations of the agents. We show that, as the population of the group increases, the results of the decentralized controller approach to its centralized counterpart with significantly lower computational cost.