Cooperative Sequential Composition Control for Compliant Manipulation

Abstract

A convoluted manipulation task involves extensive planning and the use of a supervisory controller to execute the desired task. One controller specification is generally unsuitable to perform the complete manipulation task. Manipulation involves contact of the object being manipulated with robots, surfaces and other objects in the scenario. The dynamics of a manipulation change when contact amongst these components are made or broken. Using this paradigm of contact amongst robots, objects and surfaces, the Robot Contact Language (RCL) is developed. Using a combinatory logic, the set of all possible contact combinations of the given components can be generated. Using a few simple rules such as making and breaking contacts, a “contact map” is then devised. On a symbolic level, a potential manipulation task can already be planned with the help of this map by traversing from the initial contact combination or “contact mode” to the goal contact mode. The contact map is enriched with the available geometric information (robot workspaces, surface geometry, etc.) and spatial relationships can then be defined amongst these contact modes for manipulations and mode transitions. Given the initial and final position of an object to be manipulated, planning begins with a simple graph search on the contact map for the shortest path form the initial to the final contact mode. The modes present in this path automatically divide the task into various subtasks. Manipulation planning can then be done “locally” in each of these contact modes with the aim to proceed further towards the final goal. Each of these sub-tasks can be achieved with a dedicated controller specification. A supervisory controller is needed to bring all these set of controllers together and execute them in a hybrid manner. For this purpose, the idea of sequential composition has been used. By defining the domains of attraction of each controller and the goal-sets already lying in the overlapping state-spaces, the controllers are executed sequentially to achieve the overall manipulation task. As contact involves interaction forces, a study is also done to understand the working of a spatial spring based impedance controller to achieve compliance in the robotic arm. Simulation have been conducted on Matlab and VREP software to validate the usage and reliability of manipulation planning based on contact maps as well as the performance and versatility of the compliant robotic arm.