Optimization of Motion Primitives for High-Level Motion Planning of Modular Robots

Abstract

Access to computing and storage facilities owned by parties and projects contributing to the National Grid Infrastructure MetaCentrum, provided under the programme "Projects of Large Infrastructure for Research, Development, and Innovations" (LM2010005), is greatly appreciated. Modular robots are highly flexible systems that are able to reconfigure to various shapes and to perform diverse locomotions. To visit a desired place in a complex environment, a robot can be equipped with several motion primitives that are switched using a high-level motion planning. The utilization of the high-level motion planning allows to simplify the motion primitives, which only need to provide short motions in a vicinity of the robot. This paper investigates how to automatically optimize these motion primitives of modular robots using an evolutionary approach. The proposed optimization is experimentally verified in simulated scenarios and in experiments with real robots.

Online motion planning for failure recovery of modular robotic systems

“…to cope with failures. As the online evaluation of fitness on a physical robot is time consuming, an offline optimization using physical simulation should be run first to find a good solution that can be applied to the robot [15], [7].…”

“…A fast optimization of CPG-based primitives can be achieved using Particle Swarm Optimization (PSO) [15].…”