Ropes and rope-like objects (e.g., chains, cords, lines, whips, or lassos) are comparatively cheap, simple, and useful in daily life. For a long time, humans have used such structures for manipulation tasks in a qualitatively different ways such as pulling, fastening, attaching, tying, knotting, and whipping. Nevertheless, these structures have received little attention in robotics. Because they are unactuated, such structures are regarded as difficult to model, plan and control. In this dissertation, we are interested in a mobile robot system using a flexible rope-like structure attached to its end akin to a 'tail'.
Our goal is to investigate how mobile robots can use compliant, unactuated structures for various manipulation tasks. Robots that use a tail to manipulate objects face challenges in modeling and planning of behaviors, dynamics, and combinatorial optimization. In this dissertation, we propose several methods to deal with the difficulties of modeling and planning. In addition, we solve variants of object manipulation problems wherein multiple classes of objects are to be transported by multiple cooperative robots using ropes.
Firstly, we examine motion primitives, where the primitives are designed to simplify modeling and planning issues. We explore several sets of motion primitive where each primitive contributes some aspect lacking in the others. These primitives are forward models of the system's behavior that predict the position and orientation of the object being manipulated within the workspace. Then, to solve manipulation problems, we design a planner that seeks a sequence of motion primitives by using a sampling-based motion planning approach coupled with a particle-based representation to treat error propagation of the motions. Our proposed planner is used to optimize motion sequences based on a specified preference over a set of objectives, such as execution time, navigation cost, or collision likelihood. The solutions deal with different preferences effectively, and we analyze the complementary nature of dynamic and quasi-static motions, showing that there exist regimes where transitions among them are indeed desirable, as reflected in the plans produced.
Secondly, we explore a variety of interesting primitives that result in new approaches for object manipulation problems. We examine ways two robots can join the ends of their tails so that a pair of conjoined robots can encircle objects leading to the advantage of greater towing capacity if they work cooperatively. However, individual robots possess the advantage of allowing for greater concurrency if objects are distant from one another. We solve a new manipulation problem for the scenarios of moving a collection of objects to goal locations with multiple robots that may form conjoined pairs. To maximize efficiency, the robots balance working as a tightly-knit sub-team with individual operation. We develop heuristics that give satisfactory solutions in reasonable time. The results we report include data from physical robots executing plans produced by our planner, collecting objects both by individual action and by a coupled pair operation.
We expect that our research results will help to understand how a flexible compliant appendage when added to a robot can be useful for more than just agility. The proposed techniques using simple motion models for characterizing the complicated system dynamics can be used to robotics research: motion planning, minimalist manipulators, behavior-based control, and multi-robot coordination. In addition, we expect that the proposed methods can enhance the performance of various manipulation tasks, efficient search, adaptive sampling or coverage in unknown, unstructured environments.
- Shell, Dylan Associate Professor - Term Appoint