Advancing robotic mobility through innovative locomotion systems, dynamic control algorithms, and adaptive navigation strategies
A quadruped robot that can dribble a soccer ball under the same real-world conditions as humans. The system learns to dribble through deep reinforcement learning, demonstrating whole-body coordination and dynamic balance while performing complex athletic maneuvers. The robot can navigate various terrains while maintaining ball control, showcasing advanced locomotion capabilities.
Project Website
We propose a two-stage pipeline that enables robots to perform dynamic, goal-driven tasks like lifting, throwing, and dragging with high fidelity from simulation to reality. Our Unsupervised Actuator Net (UAN) leverages real-world data to bridge the sim-to-real gap for complex actuation, while pre-training and fine-tuning strategies guide robust learning.
Project Website
We introduce a locomotion controller that learns a family of diverse walking strategies, enabling robots to rapidly adapt to new tasks and environments without retraining. This approach unlocks robust, real-time gait selection for challenges like crouching, hopping, stair climbing, and more.
Project Website
Advanced force control strategies for compliant locomotion that enable robots to adapt to varying terrain conditions and maintain stability through intelligent force distribution. Our research focuses on developing controllers that can modulate contact forces in real-time, allowing for smooth transitions between different walking surfaces and improved energy efficiency.