Saida Liu

mail · GitHub

I am an undergraduate at Kobe University, currently an exchange student at TUM (2025/26). My interests lie in safe and general physical intelligence, especially Generative models, Reinforcement Learning and Control Theory.

News

Jan 2026 — Joined the Learning Systems and Robotics (LSY) Lab at TUM as a Research Assistant.

Jan 2026 — Our paper "MATT-Diff: Multimodal Active Target Tracking by Diffusion Policy" was accepted for the L4DC 2026 conference.

Publications & Preprints

MATT-Diff: Multimodal Active Target Tracking by Diffusion Policy

Saida Liu, Nikolay Atanasov, Shumon Koga
Preprint (Submitted to L4DC 2026)

paper | code

Technical Notes & Open-source

[PDF] The Mathematical Foundations of RL from an Optimization Perspective

An in-depth note bridging reinforcement learning and constrained optimization. It covers fundamental structures including KKT conditions, and demonstrates how to achieve end-to-end safe set learning via implicit differentiation.

[GitHub] FSNet-test

A demo of FSNet, an architecture designed for differentiable constrained optimization. It implements a differentiable projection layer that unrolls optimization processes to learn policies under hard constraints.