🇳🇱 Amsterdam, Netherlands · 1d ago
PhD Candidate in Mechanism-Informed World Models and Agents
University of Amsterdam
LinkedInmidEnglish-friendly
Want to teach machines the mechanisms behind how the world changes — and build agents that act on them?Working at the UvAJoin us!World models are controllable, physics- and mechanism-grounded simulators of reality: they let an agent rehearse the consequences of its actions thousands of times before it acts in the real world. Paired with agents that plan, reason and act inside them, they are emerging as the substrate of Physical AI — and they cut across domains that look different on the surface but share the same scientific core: a robot rehearsing a manipulation task in a digital twin, and a biomedical model simulating how a tumour responds to therapy across biological scales, are both betting that structure beats brute scale.In this PhD you will develop mechanism-informed world models: models that do not merely predict the next frame, but discover the underlying events, causes and temporal relationships that drive change — and expose them as controllable, intervenable structure. The question is not only “what happens next?” but “what are the latent actions and mechanisms, and how do we tokenize, learn and steer them in a controllable manner?” This is one of the least-explored and most consequential frontiers in Physical AI.A good salary Your salary will be determined on the basis of your experience and seniority, in accordance with the Labour agreement. Your grading and therefore the amounts mentioned are based on full employment of 38 hours per week. 232 vacation hours We also offer parental leave, partner leave and social engagement leave. Through the Employment Conditions Menu Card, there are opportunities for even more vacation hours.Financial arrangements On top of your salary, you receive holiday allowance (8%) and a year-end bonus (8.3%). You build up a good pension with ABP. And in addition, we offer many other allowances and schemes.Career development and room for growth The UvA encourages you to continue to professionalise through training courses and open UvA lectures.Family matters Relocating to the Netherlands with your family? Find out about daycare, schooling, health insurance or employment options and support for your partner or spouse. Including the IDCN-network which provides support for your spouse through various training sessions, social activities, and networking events. All About This VacancyThis is what you will doYou will help define and build the next generation of world models and the agents that act in them. Concretely, you will:develop mechanism-informed world models that extract events, causes and temporal/causal relationships from observation, rather than only fitting pixels or trajectories;design ways to tokenize and learn latent actions and dynamics, and to embed controllable inductive biases into world models and agents;build agents that plan, reason and act inside these models — and that remain controllable, auditable and interpretable;validate your methods across domains: robot learning and manipulation, embodied agents, and biomedical dynamics (e.g. mechanism-informed generative models of biological processes), demonstrating that the same principles transfer;publish at top-tier venues (CVPR, ICLR, ICML, NeurIPS, ECCV) and present your work at international meetings;contribute to the life of the lab — reading groups, seminars, collaborations — and assist in teaching at the bachelor and master level.What We Ask Of YouYou are curious and rigorous; when a result is surprising, you dig until you understand the mechanism rather than settling for the metric. When experiments get messy you keep your cool, and you enjoy working in an interdisciplinary, international team.Your experience and profile:an MSc (completed before the start date) in Artificial Intelligence, Computer Science, Electrical Engineering, Physics, (Applied) Mathematics, or a closely related field;a strong foundation in machine learning / deep learning and solid programming skills (e.g. Python, PyTorch);an affinity for one or more of: world models, generative modelling, causal/representation learning, reinforcement or imitation learning, dynamical systems, or robotics;the ability to formulate and pursue research questions independently, and to communicate them clearly;a professional command of English (the working language of the lab); willingness to learn some Dutch is welcome but not required.It is a plus if you additionally have experience with robotics, physics engines or simulators, causal inference, or biomedical / scientific data — and any track record of research output (thesis, preprints, open-source code) is appreciated.This Is What We Offer YouA temporary contract for 38 hours per week for the duration of 4 years (the initial contract will be for 18 months and, after a satisfactory evaluation, extended for a total of 4 years). The preferred starting date is as soon as possible / to be discussed. This should lead to a dissertation (PhD thesis). We will draft an educational plan that includes attendance of courses and (international) meetings. We also expect you to assist in teaching undergraduate and master students.The gross monthly salary, based on 38 hours per week, ranges between € 3,059 in the first and € 3,881 in the final year (scale P), in line with the Collective Labour Agreement of Universities of the Netherlands. This does not include 8% holiday allowance and 8.3% year-end allowance. The UFO profile of PhD candidate (Promovendus) is applicable.Curious about our extensive secondary benefits package? You can read more about it on the UvA website.Your place at the UvAYou will work in this teamThe Faculty of Science has a student body of around 8,000, as well as 1,800 members of staff working in education, research or support services. Researchers and students at the Faculty of Science are fascinated by every aspect of how the world works, be it elementary particles, the birth of the universe or the functioning of the brain.You will join the CyPhai — Cyberphysical AI Lab, led by Prof. Stratis (Efstratios) Gavves, part of VISLab (Video & Image Sense Lab) at the Informatics Institute (IvI), Faculty of Science, University of Amsterdam. CyPhai pursues a single north star: algorithms that understand the dynamics of the physical world and let embodied agents act safely, reliably and accountably.The lab sits at the intersection of an unusually broad set of programmes that all converge on the same loop — reconstruct, imagine, act, reward, monitor. These span robot learning (compositional world models and robot imitation learning, with an extended collaboration with Toyota Research), the QUVA 2.0 lab with Qualcomm, the POP-AART lab with Elekta and the Netherlands Cancer Institute, and the AIRIS programme on mechanism-informed generative models for predictive and personalised medicine. CyPhai is part of the ELLIS network of excellence in AI.You will be embedded in a vibrant, international team of PhD students and postdoctoral researchers, with access to substantial GPU compute, real robots and rich biomedical and physical-world data. You will publish at the leading venues in the field — CVPR, ICLR, ICML, NeurIPS, ECCV — and present your work internationally.Want to know more about our organisation? Read more about working at the University of Amsterdam.More About The UvAThe University of Amsterdam is ambitious, creative and committed. An inspiration to students since 1632, a vanguard player in international science and a partner in innovation.The University of Amsterdam is the largest university in the Netherlands, with the broadest range of courses on offer. An intellectual hub with 42,000 students, 6,000 staff and 3,000 PhD students. Connected by a culture of curiosity.Important to knowYour application & contactIf this profile fits you and you are excited by the challenge, we look forward to receiving your application. You can apply online via the button below. We accept applications until and including 20 August 2026.Applications should inclSourced from LinkedIn. Relocantly aggregates public job postings; apply on the original site.