SPR113: Wheeled Bipedal With Locomotion & Posture Control For Using Reinforcement Learning Techniques

Andrew Ng Chee Wei ASIA PACIFIC UNIVERSITY

The main aim of this project is to develop a control system for a wheeled bipedal using Cascade PID Controller scheme that can perform autonomous tasks. The wheeled bipedal prototype design was formulated and built for testing with the controller scheme and was later tested with Nav2 to undercover its capabilities in performing autonomous navigation. The physical prototype is developed using FOC drivers with custom gearboxes as its hip actuators and outrunner motors as its wheels. The leg actuation is controlled by a 4-bar link system to reduce weight and width of the robot. The control system leverages ROS2 for rapid prototyping in software integration, where the hardware interface, odometry computation, control system, SLAM, and the autonomous navigation was developed on. The physical prototype was tested with the proposed control method’s response to height change at 0 velocity commands, response to velocity tracking at a fixed height, response to nonzero roll angles, and its performance with integration to an autonomous navigation stack. At the end, the cascade PID controller was able to self-correct itself whilst keeping track of height changes and velocity commands, and it was able to track commands from the autonomous navigation stack to perform autonomous tasks.