| Popis: |
This thesis aims to contribute on the motion planning and control problem of the locomotion of humanoid robots. For the motion planning, various methods were proposed in different levels of model dependence. First, a model free approach was proposed which utilizes linear regression to estimate the relationship between foot placement and moving velocity. The data-based feature makes it quite robust to handle modeling error and external disturbance. As a generic control philosophy, it can be applied to various robots with different gaits. To reduce the risk of collecting experimental data of model-free method, based on the simplified linear inverted pendulum model, the classic planning method of model predictive control was explored to optimize CoM trajectory with predefined foot placements or optimize them two together with respect to the ZMP constraint. Along with elaborately designed re-planning algorithm and sparse discretization of trajectories, it is fast enough to run in real time and robust enough to resist external disturbance. Thereafter, nonlinear models are utilized for motion planning by performing forward simulation iteratively following the multiple shooting method. A walking pattern is predefined to fix most of the degrees of the robot, and only one decision variable, foot placement, is left in one motion plane and therefore able to be solved in milliseconds which is sufficient to run in real time. In order to track the planned trajectories and prevent the robot from falling over, diverse control strategies were proposed according to the types of joint actuators. CoM stabilizer was designed for the robots with position-controlled joints while quasi-static Cartesian impedance control and optimization-based full body torque control were implemented for the robots with torque-controlled joints. Various scenarios were set up to demonstrate the feasibility and robustness of the proposed approaches, like walking on uneven terrain, walking with narrow feet or straight leg, push recovery and so on. |
