Decentralized collective manipulation

Go to related publications

Introduction

As a solution for manipulating relatively large and heavy objects via agents that are incapable of accomplishing the task individually, collective manipulation is one of the frequently observed behaviors in various insect colonies during the foraging process. As an impressive example, we may refer to the large prey retrieval of Eciton Burchellii (also known as army ants). Collective manipulation can be employed in a wide range of applications that include carrying and assembling parts for automated on-site construction; object manipulation and assembly in factories; search and rescue operations in disaster relief actions; and debris collection. Although utilizing a group of simple agents in comparison to a single well-instrumented agent increases manipulation dexterity, reliability, and robustness, it introduces challenges on team formation, organization, and control. In general, collective manipulation can be achieved through two fundamental control strategies: centralized and decentralized. Although the centralized control approaches, which mostly focus on group formations, can guarantee a form of optimality, they suffer from intense internal communications between the group members that makes them impractical for many real applications.

In parallel to my PhD dissertation, I am working on analysis, control and planning methods for swarm and multi-agent systems. In this regards, my colleagues and I have developed a decentralized algorithm to distribute a desired force and moment vector between the agents of a swarm team that are involved in a manipulation task. We have experimentally demonstrated that, using the proposed algorithm, a multi-robot / swarm team can robustly manipulate an object with an arbitrary shape and weight to a desired position and orientation without any need for information about team formation (number of agents and their relative position with respect to each other) or a leader-follower scheme. By not relying on the information about the team formation, the proposed method eliminates the need for any inter-agent communication and results in faster and more robust execution of the task. Furthermore, through analytical proofs and simulations, we have shown that our method is extendable to impedance control scenarios that provides a possibility of cooperation between humans and robotic teams. Currently, we are developing the required building blocks to extend the algorithm to swarm construction scenarios.

In addition to theoretical developments, we have designed and fabricated a low-cost holonomic-drive robotic platform (Δρ), to experimentally validate the performance of the proposed method. The holonomic locomotion system of Δρ allows the robot to apply forces in any planar direction. Δρ robots and some snapshots of the experimental setup are illustrated in the following figures.

Delta-rho robots Delta-rho robots carrying puzzle pieces Delta-rho robots carrying an object Delta-rho robots carrying puzzle pieces
Δρ robots and experimental setup (click to enlarge image)

A brief description of the proposed algorithm, robot designs and experimental setups are summerized in the following video (credit: Shadi Tasdighi Kalat, stasdighikalat[at]wpi[dot]edu).