Autonomous Generation and Control of Central Pattern Generator Networks for Modular Robot Locomotio

Modular robots hold the promise of being a complete solution to many problems within science. Their adaptability within hardware and software can provide the right robot for each and every situation. Properly controlling the hardware with coordinated locomotion is challenging but essential for the modules to work together and perform the assigned tasks. This research study has developed a two tiered approach which can coordinate and control the locomotion of modular robots assembled into bio-inspired shapes. The lowest level of control is in the joint space whereby all joints need to perform the right actions at the right time to produce a collaborative effort that results in locomotion. To accomplish this the Central Pattern Generator (CPG) network concept is applied from spinal vertebrate locomotion control. Each joint within the robot contributes in a small way to produce a coordinated, collaborative motion utilizing the whole body in unison. On top of the CPG network is a controller derived from the brain's cerebellum control system to modulate the CPG network to perform specific tasks such as path following.;To accomplish the CPG-based coordinate locomotion, a toolkit for automatically generating the Central Pattern Generator equations is presented. Robot shapes created in XML are read and parsed to determine sub-structures of the robot which adhere to common and known locomotion patterns for the specific modular robot currently utilized. Locomotion is based upon coordinated whole-body motion which is necessary for low-powered modular robots, such as the Linkbot used in this research, to create locomotion. By parsing the structure, the number of optimizable control parameters is drastically reduced to aid in efficient simulation of motion parameters. A simulation-based Genetic Algorithm process is used to generate motions and optimize the parameters of motion. The reduction in control parameters for large structures is an order of magnitude from the theoretical maximum. From simulated results of various shapes constructed out of the modules, the parsing of the shape increases the robot's linear speed when compared to optimizing all variables of the CPG network. The CPG networks capabilities are validated with hardware versions of the robots.;Once the CPG network has been determined which effectively controls a particular shape of robot, the network can be modulated to produce specific locomotion capabilities. Coordination between the joints is created by the network to produce locomotion. The control scheme on top of the CPG network is necessary to create a robot that can move predictably and follow objective trajectories. Again the control system takes inspiration from biology to provide real-time control of the CPG network that is controlling the modular robot configurations. In this research the brain's cerebellum, which is in charge of modulating the walking characteristics of humans, is modeled to modulate the CPG network to create path-following robots. Waypoints are used to generate cubic spline paths for the robot; a camera is used to track the robot's heading in reference to the path; and the error in heading is passed into a fuzzy controller to convert the heading error into a turn signal that is applied to the robot's CPG network. The CPG network and controller are calculated online in real-time to follow the desired trajectories. Experiments have been completed within a simulation environment built specifically for the robots used. The robots are able to accurately track the various paths laid out irregardless of the complexity and length. Root Mean Square (RMS) error is calculated for each of the experiments and shows that the robots can maintain an error between five and twelve centimeters for paths that average eight meters in length. For the snake robot the robustness of the controller is checked by adjusting the friction between the robot and the ground. This controller is not affected by changes in friction which alter the motion characteristics of the CPG network.