I am currently writing a master’s thesis on controllable multi-agent motion prediction with diffusion models and need a partner who can guide me through both the theory and the Python implementation. The core of the work is to build a diffusion-based model that predicts the future trajectories of multiple interacting agents while allowing explicit control through conditioning signals. I have already surveyed the literature, but I want to deepen my understanding of how diffusion processes are adapted for sequence forecasting, how conditioning is injected, and how training objectives differ from the image domain. On the coding side, I am working entirely in Python, leaning on PyTorch, NumPy and common scientific stacks. I would like clean, well-documented modules that I can iterate on quickly, plus explanations of each critical section so I can justify design choices in the thesis text. Deliverables • Step-by-step walkthrough of the mathematical foundations behind diffusion models for motion prediction and their controllability extensions • Modular Python code that trains, validates and tests on a standard multi-agent trajectory dataset (ETH/UCY or similar) • Inline comments and a short technical note clarifying how each block maps to the theoretical concepts discussed • A brief meeting or written summary after major milestones so I can incorporate results into my manuscript confidently If you have hands-on experience with trajectory forecasting and diffusion-based generative models, I would love to collaborate.