Biomimetic Control Applied On Robots

Over billions of years,animals have evolved extraordinary motion abilities to adapt to en-vironmental changes,which meets their needs for predation,escape,reproduction and migration.Compared with traditional artificial mechanical systems,such as vehicles,airplanes and ships,the motion of animals show stronger environmental adaptability and stability.With extensive appli-cation requirements on military reconnaissance,environment detection,disaster relief,aerospace,and medical surgery,developing a bionic robot as flexible as an animal via transplanting natural locomotion to a mechanical system has become a research hotspot in robotics fields.Inspired by the biological motion mechanism of animals,biomimetic control approaches to enable precise motion control of bionic robots is investigated in this thesis.This study has two levels,namely locomotion pattern control and body motion control.The former focuses on exploring actuator control for bionic robots interacting with environment to gen-erate desired locomotion patterns,and the latter investigates controlling body motion of bionic robots in particular locomotion patterns.The main contributions of this thesis areAccording to the similarity of the appearance and morphosis between bionic robots and cor-responding animals,the locomotion control based on general internal model,namely GIM,is proposed to regenerate natural locomotion on bionic robots.Through imitating the con-trol mechanism of animal locomotion,an artificial central pattern generator,namely CPG,is used to stimulate the GIM to ensure the coordination and stability of robots locomotion.Moreover,due to the spatiotemporal scalability and the phase shift property of the GIM,the proposed locomotion control method can effectively transfer animal locomotion into robot platforms via only tuning two parameters.Consequently,simulations and experiments on a snake-like robot prove the effectiveness of the proposed GIM-based biomimetic learning scheme on locomotion generation.2.The parameters tuning of a CPG-based locomotion controller for robots always remains a huge challenge.To overcome this challenge,this work presents a reinforcement learning-based data-driven controller design approach.Taking a hexapod robot with spatial linkage structure as a study platform,a CPG with the two-layer topology structure and the symmet-rical coupling parameters is designed,which narrows the search space of CPG parameters.Furthermore,through analyzing mechanical and kinetic constraints of the robot,a targeted reward function with the constraints is designed,and an effective deep deterministic policy gradient-based self-tuning method is proposed for tuning CPG parameters to search the best locomotion online.Finally,the feasibility of the proposed method is verified via real robot experiments in four different terrains3.To solve precise body motion control of bionic robots in known environment,inspired by biological body motion control and regulation mechanism,a cerebellum-based biomimetic hierarchical control scheme is proposed.In the proposed scheme,a CPG is designed to enable the robot to generate desired locomotion in low-level control and a cerebellum-like controller is developed to control the robot body motion under particular locomotion patterns in high-level control.Due to the online learning capability,the proposed cerebellum-like controller can compensate model uncertainties during the robot exercising and thus the control system of the robot features good robustness and environmental adaptability.Finally,compared with a traditional controller,both trajectory tracking simulation and experiment results on a snake-like robot show the performance of the proposed controller is better.4.To deal with motion planning of bionic robots in unknown environment,imitating biologi-cal iterative learning mechanism in motion planning,a variable trajectory adaptive iterative learning control,namely ILC,scheme is proposed.Through repetitive iteration and traj ectory adjustment,the robot detects boundary threshold conditions in every iteration,and then adds corrected position points into the feasible solution space of desired traj ectory according to the predefined rule.After enough iterations,the robot is capable of exploring a feasible solution from the start point to the end point and then tracks the desired trajectory via ILC.A pipe experiment on a snake-like robot is conducted,and the result proves the proposed control scheme can effectively achieve autonomous motion planning in unstructured environment.