Research And Application On Visual Terrain Classification And Gait Planning Approaches Of Quadruped Robot

The four-legged mammals have the huge advantages in environmental adaptability, kinematic dexterity and efficiency in nature, which are able to locomote in almost any ground on the earth with their legs and feet. The quadruped robot is the optimal form among the multi-legged robots considering all the factors, like manufacturing cost, controlling complexity and stability and so on, together. The quadruped robot has wide fields of applications in military-civil material transportation, field exploration and expedition, rescue mission under unstructured environment. Therefore, structuring of high performance quadruped bionic robot with the characteristics of four-legged mammal has been the dream of mankind. In fact, the research of multi-legged robots has already been developed under the continuing funds of The Defense Advanced Research Projects Agency (DARPA) since the Second World War.The robot-terrain-ground system is fundamentally a complex multi-body nonlinear rigid-flexible one with characteristics of multi-chain, moving base, variable structure, strong coupling, under-actuated, closed-chain and non-holonomic constraints. This is caused by the complex interactions between the robot and terrain-ground due to the diversity and complexity of the terrain-ground. Therefore, the research of scientific problems and intrinsic mechanism on quadruped robot and the understanding of robot basic characteristics are urgently needed. The construction of high performance quadruped robot with highly dynamics, fast speed and more payload capability is a dominant research direction and development trend. This dissertation focuses on the visual terrain classification algorithm, the design of leg structure and the configuration mode of quadruped robot, as well as the kinematics analysis and gait planning methods of quadruped robot based on improving the environmental adaptability and motion stability. The main work and contributions are summarized as follows: (1) Two key issues, the extraction approach of visual terrain feature and the fast terrain classification approach, on influencing the classification accuracy have been studied firstly in order to improve the terrain classification ability of robot. In the extraction approach of visual terrain feature, training images are convolved with the MR8filter banks. Exemplar filter responses are chosen as texton dictionary via k-means clustering. Based on the texton dictionary, the featue histogram vectors of visual terrain images are generated by means of spatial pyramid matching method. In the algorithms of terrain classification, the intelligent optimization strategies and tunable activation function are introduced in the extreme learning machine fast learning algorithm, a method of intelligent optimization algorithms of differential evolution and particle swarm optimization for Extreme Learning Machine-Radial Basis Function (ELM-RBF) neural networks and an Extreme Learning Machine with Tunable Activation Function (TAF-ELM) learning algorithm are proposed. The validity of this approach has been verified by applying the visual terrain classification approach is applied to the feature classifcitaon of terrain images based on combination of the feature extraction method of terrain images with the fast TAF-ELM learning algorithm. And thus the simulation results show that the remarkable improvement of the approach in this dissertation can improve the classification rate accuracy of terrain images with good efficiency.(2) The leg joints structure of the quadruped robot is designed based on the leg bionic structure of horse/mule. And the optimum configuration method of quadruped robot is verified by means of co-simulation ADAMS and MATLAB. The forward kinematical equation with four degrees of freedom and the inverse kinematics equation with controllable landing angle are derived, which provide the theoretical basis for improving dynamic characteristic, payload capability and environmental adaptability of the quadruped robot.(3) A trot dynamic gait planning method of quadruped robot is proposed based on summarizing the basic gait planning approaches of quadruped robot. The correctness and validity of the kinematics equation are verified in terms of the derived kinematical equation and dynamics gaint planning method from the point of both simulation and experiment. In addition, the influence of the landing angle on the locomotor performance of the quadruped robot is studied by means of ADAMS and MATLAB co-simulation.(4) Considering the properties of complex modeling and single period planning of the model-based gait planning approach, this dissertation studies the bionic gait planning approach based on the Central Pattern Generator (CPG). A new central pattern generator controller for quadruped robot is proposed based on the Wilson-Cowan neural oscillator. By replacing the gait matrix of neural oscillators, simulation results show that the presented CPG controller can generate different quadruped gaits and change its rhythmic patterns smoothly, which provide a theoretical basis for the gait planning of real quadruped robot.