چكيده به لاتين
Quadrupedal motion control is a critical issue in robotics, which, although relatively new, is one of the most important challenges in modern control systems. Quadruped robots, inspired by biological organisms, possess a high number of degrees of freedom, making their control complex. Therefore, utilizing an appropriate model capable of handling full or partial control of these robots is of paramount importance. Additionally, one of the main challenges in quadruped locomotion is developing a controller that enables the robot to traverse uneven terrains while adapting its movement to the surrounding environment.
In recent years, various controllers have been developed for quadruped robots to facilitate movement over rough terrains. One of the most effective controllers used for this purpose is the Model Predictive Controller. Along with this controller, the robot requires an accurate footstep planner capable of determining the exact placement of the robot’s feet on uneven surfaces. The core idea of this research is to utilize an MPC alongside an empirical method for footstep planning. This approach aims to optimize ground reaction forces and enable stable movement of the robot on stairs and sloped surfaces.
In this research, we extended the software developed by MIT, initially designed for flat surfaces, to enable the Mini Cheetah quadruped robot to move on stairs and sloped surfaces. We proposed two algorithms for the robot’s movement—one for slopes and one for stairs. Additionally, to reduce the need for manual parameter tuning when navigating stairs, we developed a neural network that automatically determines the position of the robot’s steps.
The results of this study demonstrate that, with the use of this controller, the Mini Cheetah robot can move on slopes of up to 25 degrees and on 12-centimeter-high stairs, without slipping, while ensuring that ground reaction force constraints are maintained.
