At Mujin Inc. in Tokyo, Japan, I had the privilege to intern and contribute to groundbreaking robotic technology. I spearheaded the development of a robot dynamics identification feature for the Mujin Controller—touted as the "world's first AI-driven intelligent robot controller for motion planning". This enhanced feature refines the robot torque model, thereby optimizing motion planning.
Specifically, it identifies and adjusts coefficients for:
Motor Friction (both viscous and coulomb)
Mass & Center of Mass (COM) for Links
Inertia Tensor
You can see the prowess of this feature in the simulation video provided, which highlights automated inertia pose validation based on real-world experiments.
Proudly, this feature transitioned from a conceptual phase to actual production and is now a staple in Mujin controllers, delivered to industry giants like PALTAC, Askul, JD.com, and more.
Further accomplishments during my tenure included:
Integration of Mujin's advanced robot test trajectory generation.
Creation of bespoke data analysis, fitting, and optimization tools and algorithms.
Enhancement of user-centric data visualization tools, tailored specifically for dynamics validation.
Short cameo of me working on dynamics identification feature @ 0:33 :)
