Collective Transport of an Unknown Object by Multi Robots with Limited Sensing
Doctor of Philosophy
This thesis presents a fully distributed approach to retrieve a large object from an unknown environment. The object is assumed to be located in an environment without GPS or Internet infrastructure. The object is too heavy to transport by one robot. The collective transport problem is broken into five major steps: 1) Exploring the unknown environment and finding the object. 2) Grasping the object. 3) Characterizing the object. 4) Planning a path to the desired location. 5) Transporting the object to the desired location. This thesis presents efficient distributed algorithms for robots with limited sensing to accomplish steps three to five. Object characterization includes centroid estimation and object dimension estimation. Two algorithms are developed for centroid estimation. In the first algorithm, each robot uses a communication tree to compute the sum of its children's positions. The second algorithm is based on pipelined consensus, which is an extension of pairwise gossip-based consensus. Two algorithms are presented to estimate object's dimensions. The first one is a distributed principal component analysis algorithm, and the second one is the distributed version of rotating calipers algorithm. A distributed path planning algorithm is presented. Robots have already been scattered across the terrain and collectively sample the obstacles in the environment. Robots use this sampling along with the estimated dimensions of the object, from above, to construct a configuration space of robots and the object. A variant of the distributed Bellman-Ford algorithm is then used to construct a shortest-path tree. A path navigation algorithm is presented to map each path segments to a distributed motion controller that can command the robots to transport the object. Four distributed motion controllers are designed including: rotation around a pivot robot, rotation in-place around an estimated centroid of the object, translation, and a combined motion of rotation and translation. Finally, a distributed recovery algorithm is presented to recover the robots efficiently and safely after collective transport. This recovery method uses k-redundant maximum-leaf spanning trees that guarantee connectivity during the recovery. All algorithms are verified through simulation as well as hardware experiments. The results are promising, and the algorithms successfully transport convex or concave objects in simulation and hardware experiments. After robots transport the object, robots are successfully recovered at home location by using the recovery algorithm. All algorithms discussed in this thesis are fully distributed, efficient, and robust to object shape and network population changes.