Autonomous Self Balancing Two Wheeler using Deep Reinforcement Learning

This research presents a novel approach, combining software and hardware, to one of the most complex challenges in robotics and autonomous systems: enabling a two-wheeler to balance and navigate on its own. Unlike traditional methods that rely on PID controllers or basic neural networks, this work introduces a novel application of the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm which is a powerful Deep Reinforcement Learning (DRL) technique, to achieve stable, responsive, and intelligent control of a twowheeled vehicle i.e., a bicycle. The two-wheeler was customdesigned in SolidWorks, 3D printed, and brought to life using a Raspberry Pi 4B, 3 DC motors, and an accelerometer sensor, all powered by a compact LiPo battery and controlled through an L293D motor driver. The model was trained in a PyBullet simulation environment, which was built using a URDF file generated from the original CAD design, ensuring a seamless bridge between the virtual and physical worlds. At the heart of this system is the TD3 algorithm, which learns to master the delicate act of balancing while also navigating towards a target. Its ability to adapt dynamically to its conditions makes it especially well-suited for the unpredictable and dynamic nature of two-wheeler motion. Through continuous exploration and exploitation, TD3 uncovers subtle patterns in the balancing behaviour and optimizes control policies that deliver a smooth, stable ride, even under varying conditions. This endeavour not only demonstrates a successful end-to-end deployment of DRL agent on real hardware but also pushes the boundaries of autonomous vehicle research. It proves that with the right algorithms and system design, even highly unstable platforms like two-wheelers can be controlled autonomously with precision and reliability. In doing so, it opens new possibilities for the future of personal transport, robotics, and intelligent mobility. All this has been achieved at a cost less than INR 7000.