人型機器人節能步態生成器

Abstract

none

Energy efficiency is the main issue in the robotics field, especially in the humanoid robot, due to the limited power source from the battery. Efficient power consumption becomes the primary role of increasing the durability of the robot. In the humanoid robot, the main electric load is on the joint actuators. Therefore, for reducing the energy consumption, it can be formulated through gait optimization, which is selected from the optimal values of parameterized of the gait engine. This thesis proposed a method for generating a stable and energy-efficient gait for the humanoid robot that can be applied in variable speed and omnidirectional walk. The gait pattern is generated by Zero Moment Point (ZMP) preview controller and Bezier function. Gait engine is parameterized by parameters to adjust the Centre of Mass (CoM) height, body posture, and walking speed. The Covariance Matrix Adaptation Evolution Strategies (CMA-ES) has been proposed to find the optimal values that yielded a stable and energy-efficient gait in a safe simulation environment. The optimal gait parameters were verified in the simulation and real robot, able to reduce energy about 29.813 % and improve stability 20 % during training. Verification in the real robot validated the result, which can save energy about 19.905 % compared to non-optimized gait. Moreover, the optimal parameters are generalized that can be applied to variable speed and omnidirectional walk without unstable issues.