Reinforcement Learning Legged Robot
Spider-Bot is a four-legged robot that learns to walk through reinforcement learning instead of hand-tuned gait equations. Each leg carries multiple degrees of freedom in a biologically inspired layout, giving the controller enough range of motion to adapt to different terrain.
I trained it in PyBullet, letting the controller discover stable locomotion patterns through trial and error rather than scripting them by hand.
The frame mixes CAD-designed parts with FEA-checked joints, fabricated through a combination of additive manufacturing and laser cutting. A Raspberry Pi sits at the center of the electronics, driving the servo motors and controllers that move all four legs in sync.
I used PyBullet to simulate and tune gait patterns before they touched hardware, then wrote the Python scripts that generate and execute them on the real robot. The hardest part wasn't the simulation, it was getting gaits that looked stable in PyBullet to hold up once weight, friction, and motor backlash entered the picture. Two gaits made the cut: a crawl/amble pattern for steady terrain and a creep gait for more cautious footing, together carrying the robot at speeds up to 20 cm/s.