Interested applicants are invited to apply directly at the NUS Career Portal. Please note your application will only be processed if you apply via NUS Career Portal.
NUS Career Portal link: https://careers.nus.edu.sg/job/Research-Fellow-%28Smart-Systems-Institute%29/33052-en_GB/st=11B80EC9332E5DA28A92640F3B8008AC42653B22
We regret that only shortlisted candidates will be notified.
Job Description
The Smart Systems Institute (SSI) at the National University of Singapore is an interdisciplinary research institute and an experimental playground for human-centered AI technologies. Our vision is to enable situated AI assistance for every human to work, play, and learn. The research at SSI spans the spectrum of embodied AI, robotics, interaction design, augmented/virtual reality. The institute brings together researchers from diverse backgrounds, including AI, engineering, design, social sciences, to collaborate and to shape the future of AI.
We are currently looking for a Research Fellow. This position will be developing theory and engineering principles for data colleaction for training large robot policies.
Qualifications and Requirements
Requirements
- PhD in robotics, machine learning, computer vision, or a closely related field (or equivalent industry research output).
- First-author publications at top robotics or ML venues (CoRL, RSS, ICRA, IROS, RA-L, NeurIPS, ICML, ICLR).
- Hands-on experience training visuomotor policies - imitation learning, behavior cloning, diffusion policies, RL, or VLA-style models - and running them on real hardware.
- Demonstrated work on ..data-efficient learning..: extracting more from fewer demonstrations, transfer across tasks/embodiments, knowledge distillation, or pretraining-from-data approaches.
- Real-robot fluency on manipulation systems: ROS, MoveIt, motion planning, gripper/tool integration, calibration, and comfort with the failure modes of physical experiments.
- Experience designing and running data collection at non-trivial scale - teleoperation interfaces, scripted policies, or autonomous collection - and reasoning about which data is actually informative.
- Comfort with bimanual or dexterous manipulation, including contact-rich or deformable-object tasks (cloth, articulated objects, tools) where the data problem is hardest.
Strongly preferred
- Track record contributing to or scaling robot datasets (Open-X / RT-X / DROID-style efforts, or comparable internal programs).
- Synthetic data and sim-to-real: dataset generation in simulation (Isaac, MuJoCo, etc.), domain randomization, sim2real transfer for perception or policy.
- Long-horizon manipulation and loco-manipulation experience, including on legged platforms.
- Representation learning for control
