Genetic algorithm based mass distribution optimization of quadruped leg robot for walking performance enhancement

In previous research works, legged robots are typically induce stable locomotion in active compliance and passive compliance approaches. In order to realize active compliance locomotion, the detail studies on robot hardware and environmental factors are required. Hence, a controller is designed to tailor for specific environment. Therefore, designing an all-rounded controller is extremely challenging. On the contrary, passive compliance locomotion relies on the advantage of its own body. Passive compliance mechanism that currently adopted in legged robot such as spring-damper mechanisms, flexible links and components, and adjustable joint stiffness are used to store and release the impact from the environment. Therefore, stable locomotion can be acquired. However, morphological Meffect on locomotion is an important issue to study. In this thesis, the study is focused on another approach that utilize the advantage of the robot body to achieve stable locomotion, which is mass distribution of the robot. Genetic algorithm is used to search for the optimal mass distribution that carried out the farthest walking distance on various terrains. In this experiment, quadruped is used to perform repetition tests in simulated environment. In the conditions of fixed walking cycle, preset walking pattern and limited torque generation in actuators, genetic algorithm is adopted to optimize the walking distance of the robot by varying the masses of torso, upper limb and lower limb, ranging from 0.01kg to 5kg. The predefined joint trajectories are generated using Matsuoka neural oscillator network as the central pattern generator of the quadruped. Open Dynamics Engine is adopted for legged robot simulation. The robot is programmed to walk on flat terrain, inclined terrains of 0.1 radian and 0.15 radian, and declined terrains of 0.1 radian and 0.15 radian with different mass distribution, in which the value of the masses (torso, upper limb, and lower limb) are stored in the chromosomes. Thus, it allows genetic operators to take control of the information for optimization. According to the experiment results, it shows that the mass distribution of the robot substantially affect the walking distance of the robot. It also demonstrates that genetic algorithm successfully implemented to enhance walking performance by maximizing the walking distance in prefix conditions. It also leads to a conclusion that intelligently manipulation of mass distribution can extend walking distance significantly.

Preview PDF ITMA 2013 5R.pdf Download (837kB) | Preview

Actions (login required)

View Item