Research On Motion Control Based On Chaotic Neuron For Modular Self-reconfigurable Robot

Modular self-reconfigurable robots consist of several groups of basicisomorphic modules which are interchangeable with each other on structure andfunction. The robot can change connected topology or joint morphology betweenmodules depending on the tasks and operating environment to reconfigurate thetarget configuration. Compared with conventional robots, modular robot has thecharacteristics of diversity on configuration growth or locomotion mode andreconfigurability of target configuration. The method of motion planning andcontrol for fixed configuration robots is difficult to be applied. It is important toexplore a novel and facilitated motion control strategy, by which the unified motioncontrol between multiple configurations could be achieved and the abilities ofmulti-mode locomotion and configuration conversion could be aroused.Considering the mutative and complex environment, the research on self-adaptionof modular robots is significant.Artificial neural network is one of the effective methods to solve nonlinearproblems. The motion control method based on chaotic neuron bas been applied inthis paper to solve multi-mode motion problems between different configurations.chaotic CPG unit which has a chaotic solution has been designed and the delayedfeedback chaos control method has been applied to stabilize this chaotic solution tothe designated periodical orbits to activate the rhythmic signals; Other functionalunits have been introduced such as environmental preprocessing unit, velocityregulating network and phase switching network. Combined with the chaotic CPG,the motion control neural networks for modular robots has been constituted. By themotion control neural networks designed in this paper, The motion planning forseveral typical configuration such as worm-type, cross-type, and multi-leggedconfigurations has been researched to solve the control problems on morphologicaladjustment, omnidirectional motion, gait generation and etc.. The controlparameters in the model are simple and adjustable, and could stimulate diverselocomotion mode and configurations.In this paper, the sensing system has been designed for UBot robot, meanwhile,the environmental preprocessing network has been proposed to composeenvironment-motor motion control systems combined with other functionalnetworks. The simulations about getting through the tunnels, self-recovery fromunstable state, obstacle avoidance, stepping across the obstacle and dangeravoidance have been verified to achieve the environmental self-adaptive locomotion. By changing the topological connection or the joints morphology, the modularrobots could reconstruct the structure and shape to generate a wealth of locomotionconfigurations. This paper describes the basic principles of configurationtransformation, and for the configuration transformation induced by joint initialmorphological conversion, the deformation controller and motor neurons have beenestablished to achieve unified locomotion control. On the other hand, for theconfiguration transformation induced by topology structure conversion, the derivedconfigurations generated by robots deformation have been analyzed and thealgorithms for coordinated locomotion control of complex configurationtransformation have been proposed.The Ubot physical modules consist of the two rotated freedom degree joint andtwo L-type parts. In this paper, on the basis of mechanical systems, the controlsystem on processing capability, local communication, energy management andorientation identification has been improved to provide effective hardware supportfor motion control studies. In this paper, the physical experimental platform forUbot robot system has been established and the experiments of locomotion forworm-type and cross-type configurations have been finished to verify the motioncontrol neural networks proposed.