PhD in Physics-Informed Foundation Models for Robotics

About the Role

TU Delft’s Faculty of Mechanical Engineering and Systems Design seeks a PhD candidate to develop physics-informed foundation models that allow robots to generalise across tasks without extensive retraining. The position sits within a collaborative project funded by the Dutch Research Council (NWO). Employment is for four years under the Dutch university collective labour agreement, with a go/no-go decision at the end of year one.

Gross salary starts at EUR 2,872 per month in year one and rises to EUR 3,670 in year four.

Responsibilities

• Develop foundation models that incorporate physical constraints (rigid body dynamics, contact mechanics) into neural network architectures
• Evaluate models on real robotic hardware in TU Delft’s robotics labs
• Collaborate with partner groups in computer science and mechanical engineering
• Publish in top robotics and machine learning venues (ICRA, CoRL, NeurIPS, ICLR)
• Contribute to teaching assistant duties up to 20% of time

Requirements

• Masters degree in robotics, computer science, mechanical engineering, or applied mathematics, completed by the start date
• Strong background in at least two of: robot dynamics, machine learning, numerical simulation, or control theory
• Programming proficiency in Python and one systems language (C++ or Julia)
• Good command of English in writing and speech; Dutch is not required

Nice to Have

• Experience with ROS 2 or comparable robotics middleware
• Familiarity with physics simulators (MuJoCo, Isaac Sim, PyBullet)
• At least one peer-reviewed publication or strong first-author preprint
• Background in Bayesian methods or Gaussian processes

How to Apply

Apply online through the TU Delft vacancies portal by 9 June 2026. Attach a CV, a one-page motivation letter, contact details for two referees, and a copy of your Masters thesis or the most relevant writing sample. Shortlisted candidates will be invited to a virtual interview in the first week of July.