CART POLE SYSTEM ANALYSIS AND CONTROL USING MACHINE LEARNING ALGORITHMS

Abstract

The cart and pole system balancing is a classical benchmark problem in control theory which is also referred as inverted pendulum. It is a prototype laboratory model of an unstable mechanical system. It is mainly used to model the control problems of rockets and missiles in the initial stages of their launch. This system represents an unstable system because an external force is required to keep the pendulum in vertically upright position when cart moves on horizontal track. Designing optimal controllers for the Cart and pole system is a challenging and complex problem as it is an inherently nonlinear system. The principal advantage of reinforcement learning (RL) is its ability to learn from the interaction with the environment and provide an optimal control strategy. In this project, RL is explored in the context of control of the benchmark cart-pole dynamical system. RL algorithms such as Q-Learning, SARSA, and value-function approximation applied to Q-Learning are implemented in this context. By using a fixed Force value of +10N or -10N, decided by a policy that maximizes the approximate value function, the agent achieves optimal control of the system.