Tri-Bot — A Self-Taught Walking Robot Built for a Quarter of the Price

Conventional wisdom in legged robotics says four legs are the minimum for stable locomotion. My thesis challenged that assumption directly. By removing one leg and training a reinforcement learning controller from scratch, I built a robot that learns to compensate for its asymmetry and develops a functional gait through trial and error — no hand-coded motion planning. The robot is fully 3D printed, runs on a Raspberry Pi, and uses a custom ROS-based control stack. Total bill of materials: 75% cheaper than comparable quadrupeds. Locomotion quality: surprisingly good.