Research

My research focuses on reliable embodied autonomy for robots operating in cluttered, dynamic, human-centered environments.

Integrated embodied multi-agent task and motion planning (2025–Present)

• Built an integrated multi-agent framework that combines high-level task decomposition with learning-based navigation and safety-guaranteed coordination.
• Developed deadlock detection and an interface between decentralized RL controllers and a local Multi-Agent Path Finding (MAPF) solver to guarantee progress in hard cases.

Human behavior-informed safe motion planning (2024–2025)

• Developed a human action prediction module that combines human characteristic probing with forward reachability analysis.
• Designed an MPC planner that embeds uncertainty-aware safety constraints for human–robot interaction scenarios.

Human trajectory prediction for interaction (2024)

• Trained deep models to forecast human motion and integrated predictions into robot path planning for safer interactions.

Transfer learning in deep reinforcement learning (2023–2024)

• Studied generalization failures under visual/dynamics shifts and improved adaptation with self-supervised auxiliary objectives.