Legged Robotics

The goal of this project was to study and implement state-of-the-art locomotion algorithms for legged robots, focusing on both bipedal and quadrupedal systems. Our work explored approaches ranging from analytical trajectory planning to bio-inspired oscillators and deep reinforcement learning, with the aim of understanding and improving how robots achieve stable and adaptive movement.

Bipedal Locomotion

For bipeds, we implemented locomotion based on the Divergent Component of Motion (DCM) framework:

• DCM Planning – generating stable Center of Mass (CoM) trajectories.
• Foot Trajectory Planning – ensuring dynamic balance with fixed-step positioning.
• Inverse Kinematics – mapping trajectories into executable joint motions.

This resulted in a robust controller capable of maintaining balance and walking smoothly on flat terrain.

Quadrupedal Locomotion

For quadrupeds, we combined bio-inspired control with learning-based methods:

• Central Pattern Generators (CPGs) – oscillator-based gait generation (walk, trot, pace, bound).
• PD Control – Cartesian and joint-space foot placement.
• Deep Reinforcement Learning (DRL) – training policies with PPO and SAC to adapt locomotion across challenging terrains.

These approaches allowed us to explore the trade-offs between model-based control and data-driven learning, while building a foundation for more advanced legged robotics research.

Github here.

Youtube here.