Bipedal Robot

This bipedal robot project began as an intensive holiday endeavor, sparked by my fascination with a specific type of joint mechanism that enables rotation across all three axes - a perfect foundation for the hip joints of a humanoid robot.

The development process started with detailed 3D modeling in Fusion 360, where I designed a robot structure optimized around these multi-axis joints. The implementation utilized servo motors to control the various articulation points throughout the robot's frame.

What distinguishes this project technically is the reinforcement learning approach I employed for teaching the robot to balance. Using Unity's simulation environment, I created a simplified digital version of the robot and implemented a learning system where: 1. Initial iterations performed completely random movements 2. Algorithms rewarded configurations that maintained upright posture for the longest duration 3. Energy efficiency was prioritized by penalizing excessive motor jittering 4. Successful movement patterns were progressively refined through thousands of iterations

Once the virtual training yielded promising results, I 3D printed the physical components and assembled the robot with its servo motor control system. While I successfully implemented the innovative joint mechanisms as designed, the project faced challenges with the durability of 3D printed parts. The holiday period ended before I could fully address these structural limitations.

In retrospect, the primary lesson was material selection - the 3D printed components proved too fragile for the mechanical stresses involved. A future iteration would benefit from stronger materials and a more realistic timeline. Despite remaining incomplete, the project provided valuable hands-on experience with reinforcement learning applied to physical robotics, making it one of my favorite technical explorations.