Optimal learning and surface identification for terrestrial soft robots

Abstract

Soft material robots have potential for deployment in dynamic environments, e.g. search and rescue operations, owing to their impact resistance and adaptability. However, these advantages are accompanied by challenges of robot control and surface identification. The continuum, soft material robot body interacts uniquely with different environments e.g. a smooth table or a rough carpet. These interactions with the surface can be discretized and modeled using graph theory. This representation allows the robot to learn from its surroundings and generate environment-specific locomotion control sequences. Here, simple cycles of individual graphs are analogous to periodic locomotion gaits of the soft robot. Inversely, provided the knowledge of different environments (captured in the individual graphs), the robot has ability to optimally identify the environment through experimentation and interaction. This paper presents a method for soft robots to a) optimally learn the environment and b) determine optimized movements for identifying the surface of locomotion by utilizing the information from previously experienced environments. The optimized movements are identified as arcs, paths and simple cycles that yield the most contrasting costs. The surface identification is performed by analyzing the locomotion cost differential between the experienced surface interaction and that of a previously known environment. The learning and control algorithms (Eulerian path, simple cycles) are ‘arc-centric’ i.e. focus on traversing arcs. Whereas surface identification algorithms are ‘node-centric’ i.e. focus on traversing nodes (simple paths).