| Legged robot is a class of multi-dimensions and strong nonlinearity mechanical system.In addition,the characteristic of under-actuation makes the control of legged robot is a challenging task to be resolved.Control schemes for legged robot used in the past decades includes joint decoupling control,operation space position control,operation space force control and inverse dynamics control and so on.Legged robot with joint decoupling control couldn’t be adaptive to the unstructured environment because of its high stiffness,while operation space whole-body control and inverse dynamics control methods are unfit for real-time control because of high dimension,complicated kinematics and dynamics,and suffer much computation burden.In this thesis,the research on whole body control framework,optimization based torso control and single leg control,and joint actuator control is carried out on the quadruped robot prototype named “Cavalry” developed by our laboratory.The main contents and contributions of the thesis are as below,1、Framework of whole body control for legged robot.To avoid the high dimension problem existing in the full state whole body control,the whole body control framework proposed in this thesis is divided into three hierarchies including torso level,single leg level and joint level.The torso level outputs the desired contact forces between the foot and the ground based on the desired motion of the torso.The single leg level outputs the desired joint torques,positions and velocities based on the desired motion of the single leg and the desired contact forces obtained from the torso level.The joint level outputs the input voltages or currents of the actuators based on the desired joint torques,positions and velocities obtained from the single level.The hierarchical framework proposed in this thesis reduces the dimension of the control algorithm successfully compared to the full state whole body control method.2、Optimization of the distribution of contact forces.The distribution of the contact forces is optimized on the basis of the desired motion of the torso.Several optimization criteria are designed in this thesis such as minimizing the internal forces between the feet of the robot,minimizing the tangential contact forces to decrease sliding,and minimizing the tracking error in comparison with desired values.Then a standard constrained time-varying QP problem about contact forces combined with torso dynamics,unilateral and friction cone constraints is formulated,followed by an online active set time-varying QP solver.At last a smooth switch process is design to improve the compliance when the phase of a single leg changes between swing phase and support phase.3、Optimization of the distribution of joint torques and velocities of the single leg.The distribution of joint torques and velocities is optimized on the basis of the desired motion of the single leg and the desired contact forces.A new optimization criterion minimizing the sum of square of joint torques and velocities simultaneously and taking the characteristic of both swing phase and support phase into consideration is proposed.The proposed optimization criterion is converted to another optimization criterion on the joint acceleration level,and then a standard constrained time varying QP problem is formulated combined with single leg kinematics and joint physical constraints.The improved LVI-PDNN(linear variational inequality based primal dual neural network)named “CLVI-PDNN”(closed-loop LVI-PDNN)and “Mid-Value CLVI-PDNN” are proposed in this thesis to solve the time vary QP problem.At last,a kind of height control method based on stable limit cycles for hopping motion is presented.4、Simulation and experiments.The proposed whole body control framework for legged robot including optimization of distribution of contact forces,optimization of distribution of joint torques and velocities of single leg,and control of joint actuator is tested on the simulation platform in Matlab/RecurDyn simulation environment and the single leg experiment platform developed by our laboratory.The simulation results verify the efficacy of the whole body control framework and the experiments operated on the single leg platform indicate the effective control performance of the control framework in the torso height control.Experiments on quadruped robot developed by our laboratory verify the effectiveness of the constraints redundancy resolution based control algorithm proposed in this paper. |
