ZMP Analysis & Motion Generation for Humanoids

Final year project, extended

Reproduced CLoSD and added zero-moment point analysis to check whether diffusion-generated humanoid motions are dynamically stable.

Reproduced CLoSD (Tevet et al.), a pipeline that pairs a diffusion-based motion planner with a physics-simulated tracking controller, generating humanoid motions from text prompts (e.g., "walk forward").
Developed zero-moment point (ZMP) analysis to evaluate the dynamic stability of diffusion-generated motions, computing and visualizing ZMP trajectories against the support polygon.
Team project with Omer Amin, Noor Ul Eman, and Eaman Safdar. Advisor: Dr. Yasar Ayaz; co-advisors: Dr. Mohsin Kamal and Dr. Khawaja Fahad Iqbal.