BridgeDP Robotics Co., Ltd.
About BridgeDP
BridgeDP is a leading developer of the general-purpose robot "cerebellum" (motion control), dedicated to solving key technical challenges in behavior and motion control for humanoid, quadruped, and wheeled robots, while providing precise, flexible, and efficient motion control solutions for clients.
Core product: BridgeDP Engine, based on high-precision motion capture data and reinforcement learning algorithms, enables humanoid robots to perform human-like motions across core scenarios such as perception planning and legged motion control.
Role & Responsibilities
- Situation: BridgeDP provides customized motion control algorithms for multiple robot manufacturers. Client robots vary in form, including humanoid, bipedal, and quadruped platforms, and the algorithms must adapt to different hardware platforms and scenario requirements.
- Task: As a Robot Control Engineer, responsible for customized development of motion control algorithms for client humanoid robots, covering the full process from policy training to real-world deployment.
- Action: Combine Reinforcement Learning (RL) and Imitation Learning (IL) to design and train motion control policies for specific robot forms and task requirements; deploy simulation policies to real robots through Sim2Real transfer; continuously optimize policy performance to satisfy client scenario requirements; refactor the deployment code workflow to significantly improve development efficiency, reducing single deployment code development time from about one day to 1 to 2 hours.
- Key Technical Contribution: Implemented a universal solver for robot parallel structures, enabling bidirectional conversion of position, velocity, and torque between motor space and joint space across all supported parallel mechanisms.
- Result: Details coming soon, currently contributing to multiple client projects.
🎯 RoboCraft AI Productization & Standardization
- Situation: Motion control development and deployment depended heavily on specialist expertise across reinforcement learning, simulation, Sim2Real transfer, and hardware integration. The resulting workflows were difficult to reuse across robot platforms and inaccessible to many downstream developers.
- Task: Served as Product Manager for RoboCraft AI, translating the development and usage logic of motion control algorithms into a standardized product for robot manufacturers and application developers.
- Action: Mapped and abstracted the end-to-end workflow—from robot onboarding, model validation, policy training, and deployment to motion acquisition, choreography, multi-robot deployment, and asset management—and converted expert practices into reusable product modules, user flows, and standardized processes.
- Result: Helped shape RoboCraft AI into a general-purpose robot motion capability development platform, enabling motion control expertise to be delivered and reused through a unified product workflow.