Nanyang Technological University — MSc (research), Mechanical Engineering · Aug 2025 – expected Feb 2027
School of Mechanical and Aerospace Engineering, Intelligent Cybernetics Lab; advisor: Mir Feroskhan
Feiyang Song宋飞扬
Robot learning and control — making learned policies accurate, stable, and responsive on real robots.
Research master's student, School of Mechanical and Aerospace Engineering, Nanyang Technological University
Intelligent Cybernetics Lab (ICL), advised by Prof. Mir Feroskhan · Expected Feb 2027
I study robot learning and control, with an emphasis on the accuracy, stability, and responsiveness of learned policies in real-world manipulation. My current research focuses on vision–language–action (VLA) models and uses real-robot experiments to examine how observation information, action representations, and execution mechanisms affect task performance. I built the real-robot research platform of our lab end to end and carried the real-robot work of all collaborations in it. I am applying for PhD positions starting Fall 2027.
News
Oct 2026LiFT is accepted to NeurIPS 2026 as a poster.
Publications
Real-robot demonstrations (supplementary video) — click to play.
LiFT: Likelihood-Free Tree-Structured Policy Optimization for Flow-Based VLAs
NeurIPS 2026 poster · accepted
Real-robot demonstrations — click to play.
RL-APT: Reinforcement Learning for Action-Prior Transport via Generalized Schrödinger Bridges
Manuscript 2026
Under review: two additional manuscripts at ICRA 2027 (one first-author). Titles are withheld during double-blind review and will appear here after decisions.
Research Experience
- Independently built and adapted the lab's Piper real-robot research platform end to end: robot and sensor integration, teleoperation and data collection, model fine-tuning, inference deployment, action-execution optimization, and real-robot evaluation.
- First-author project (under review): independently led method design and model improvements, sensor integration on real hardware, the data pipeline, and all training and evaluation (150 demonstrations, 540 primary real-robot trials), and wrote the full initial manuscript.
- Carried out all real-robot deployment and experimental validation of VLA models for two collaborative projects, including the NeurIPS 2026 paper above: runtime and action-execution optimization, teleoperation dataset collection, and demo recording.
Education
Hohai University — BEng, Mechanical Engineering · Sep 2021 – Jun 2025
Awards & Service
- Two First Prizes, 2023 RoboCup China Open (FIRA small-size semi-autonomous events).
- Assisting paper reviewing for ICRA 2027 (ongoing).