Decentralized Motion Planning for Load Carrying and Manipulating by a Robotic Pack

In many cases, a pack of robots holds an advantage over a single robot such when an oversized or over-weighted load is to be carried. In such cases, a single robot will not do. Nevertheless, this may not be an easy task for a pack of robots as well, especially when the load needs to be lifted off the ground making the cooperative task less tolerant of errors. The limited research on such a load can be attributed to the mechanical complexity of the problem. Notably, previous studies have not considered the spatial, decentralized, communication-free scenario. We, therefore, consider a robotic pack of six agents that assumes the task of spatially moving a load through a cluttered space. As it transports the load, the pack carefully avoids planar obstacles while maintaining its orientation. To do so, we model the whole system as a six Prismatic-Prismatic-Spherical-Spherical (6-PPSS) redundant mobile platform, having twelve degrees of freedom. This paper focuses on a decentralized control scheme where no mutual communication is needed. Each agent calculates its ego movements according to the height of its corresponding load-node; the surrounding obstacles, and; the goal’s relative position. To avoid numerical errors appearing in the vicinity of singular configurations, we calculate the platform’s forward kinematics in the model’s full configuration space. We then show how this rationale can be further extended to formulate a distributed control scheme for the motion planner. We demonstrate our algorithms in several simulated scenarios and in a set of real-world experiments using specially designed omnidirectional robot agents. We test the ability of the pack to maintain the load’s orientation just by measuring the load’s height at the holding node of each agent. Lastly, we measured the time required for the pack to assume a desired load orientation. Results indicated that even in the presence of a 10-degree tilt error, the load was able to be restabilized within a maximum of 15 seconds in simulated conditions and 20 seconds in real-life experiments.