Movement of Modular Hydraulic Propulsion robots: Decentralised and reactive pose control strategies

Modular robotic systems consist of a set of reconfigurable units, called modules, which can be combined in a multitude of ways to produce robots of different shaps . One of the challenges in the design of these system is to enable them to perform precise movements in their environment. Control strategies that are centralised or rely on external sensing can limit the robustness and scalability of the system. This thesis focuses on the development of control strategies that allow the position and orientation (pose) of a modular robot to be controlled in a fully decentralised and reactive manner. The strategies are designed for the Modular Hydraulic Propulsion (MHP) system, wich operates in a liquid environment. An MHP robot is made of cubic modules, which create a fluid network when connected together. To move, the robot routes through this network fluid from the environment. A physical implementation of the MHP concept is designed, built and validated. An MHP robot’s ability to translate efficiently towards a goal is tested using occlusion based controllers, both with and without communication between modules. The robot is shown to reach the goal using either of the controllers. When using communication, an average of 70% of energy is saved, at the cost of a longer completion time. This thesis proposes multiple minimalistic controllers to control the pose of MHP robots. The robot is required to reach a goal in a preferred orientation. All of the controllers use binary sensing and actuation, with each module using only two bits of sensory information per face. The controllers are proposed for robots moving in 2D and 3D space, and use up to five bits of communication between modules. We prove that robots of convex shape are guaranteed to complete the task. Using computer simulations, the controllers are tested in different environments, using multiple module sizes and under the effect of noise. Additionally, their performance is compared against a centralised controller from the literature. Given the simplicity of the solutions, modules could potentially be realised at scales below a millimetre-cube, where robots of high spatial resolution could perform accurate movements in liquid environments.