Swing up pendubot with a GA-tuned bang-bang controller

This paper presents a new control method to swing up pendubot (pendulum robot), a typical two-link under-actuated system. The pendubot is composed of the first active joint and the second passive joint. The swing up is a complex problem to move the two links from low to top position by appropriate external torque on the first joint. A bang-bang controller is developed to apply constant output torque on the system periodically. When the two links are located about the natural low position, the constant torque is applied on the first actuated joint to make the links swing up by one or more times, finally close to upright position, the attractive basin of a linear controller for balancing. The balancing controller is designed with LQR method. The constant output torque with control basin of the bang-bang controller is determined with a modified genetic algorithm. The control parameters are verified to swing up the pendubot through simulation results with different typical experimental setups. The results indicate that a large output torque cause the two links swinging up quickly to the attractive basin, while a small torque needs several swings back and forth to reach the attractive basin, but it provides a valuable control strategy for real applications.