Research On Coordinated Locomotion Planning And Evolution For Modular Self-reconfigurable Robots

Modular self-reconfigurable robots are composed of basic modules which have certain motion and sensing abilities. It could form various configurations(morphology) to better adapt to different environments and tasks. For instance, the robot could get through a narrow hole using snake- or worm- configuration, change into a quadruped robot to across rough terrains, change into a loop configuration to roll on the flat ground with high-speed. Composed configurations are hyper-redundant because each module has locomotive joints,. These features make the planning and control for modular robots a challenging field. Coordinated motion is a basic capability of modular robots, and the key point for tackling this problem is to find an automatic locomotion generation method in an acceptable time period for an arbitrary configuration under certain environments and tasks.The basic module structure can directly affect locomotion patterns and capability. In this paper, coordinated locomotion for UBot modular robot system is investigated. UBot system has two kinds of modules, i.e., active module and passive module. The active module has the ability of connecting and disconnecting with passive modules by a mechanical claw. Each module has two degrees of freedom arranged on a L-shaped axis. To automatically generate locomotion, several key technologies should be resovled. On software part, the motion planning platform which could fast generate locomotion for different configurations is nessessary. Thus we develop a 3D dynamics simulation platform that could efficiently evolve locomotion through locomotion simulation and evolutionary algorithms. It provides an effective support for quickly generating locomotion for modular robots.A scalable locomotion controller that could adapt to changeable number of modules is an important aspect of automatically locomotion generation. In this paper, we make a systematic investigation on locomotion planning of caterpillar-like configuration and loop configuration based on serpenoid curv e and proposed serpenoid polygon model, and achieve a unified planning method for multi-mode locomotion of chain configuration with different module number. Based on that, several novel locomotion gaits are found, e.g. earthworm-like locomotion and serpenoid polygon rolling pattern. These motion patterns provide the movement basis for chained robot to adapt environments and tasks.In case of non-chain configurations, how to automatically generate locomotion. In this paper, off-line optimization that combines simulated locomotion evaluation and optimization algorithm is empolyed, which is frequently used in evolutionay robotics field. First, setting the controller expressions and parameters to be optimized. Then, utilizing fitness value obtained from simulation to guide the parameter search process. This way, we could acquire satisfied locomotion results in a limited time. For searching multiple locomotion patterns, we propose a particle swarm optimization algorithm which use behavior sparseness as the criteria metric of evolution. This means searching target is abandoned. But the results show that it can not only search the locomotion gaits with high-speed, but also can find many other motion patterns in the same time. Locomotion evolution for worm-like robot, snake robot, cross robot and quadruped robot is implemented, and simulation results demonstrate the effectiveness of this algorithm.The controller expressions of studies metioned above are predefined. That requires the designer has a deeper knowledge of the robot configurations and its geometrical characteristic. To automate this procedure, self-modeling for arbitrary configuration is investigated. In this paper, we propose a theory and framework for self-modeling motion evolution of moudar robots, it could automatically generate controller expressions and its optimized parameters in the same time. The framework include topology analysis, controller expression generation of functional sub-structure, constraint of isomorphic sub-structures, and paramenter evolution. We investigate several configurations of different morphology. The results demonstrate that self-modeling evolution of locomotion could find fast regular locomotion gait in a short time. This study provides the ability for modular robot to quickly adapt environments for changed configuration.To verify the effectiveness of planned and evolved motion results, we construct an integrated software and hardware system for UBot modular robot. A configuration recognization and gait matching algorithm bas ed on module pose list is proposed and integrated into UBot Sim. With the gait control both in virtual and physical robots, a synchronous control function ’what you see is what you get’(WYSIYG) is realized, which provides a powerful and convenient tool to test evolved gait. Multi-mode locomotions for 8-module caterpillar configuation and 12-module loop configuration are implemented to verify the scalable controller planning. Evolved gaits based on parameter optimization and self-modeling evolution are also implemented. These experiments verify the effectiveness of evolved locomotion in developed modular robotic simulator — UBot Sim.