Effectively tuning a PID controller can be difficult without prior experience or knowledge of the system being controlled. Reinforcement learning is a tool that allows automatic PID tuning with adaptability to environmental change. This technique was utilised for a single degree-of-freedom robot designed for human interaction, proving the validity of the TD3PG algorithm for reference tracking and rehabilitation exercises. These results were measured by the root mean square error of the system and compared to a classical PID controller to determine whether the adaptability improved the system tracking ability. Results showed the classical PID controller resulted in smaller RMSE measurements for a multitude of input signals including sine waves and multi-step functions when the environment remained constant. The adaptive PID controller resulted in smaller RMSE measurements for all input signals when the environment changed to reduce the amount of torque applied to the plant, representing a motor power failure. It is believed that a classic PID controller is better suited for systems with low input frequency and low system uncertainty while adaptive PID controllers are better for systems with changing environments or input signals.