PhD Position: Safe Planning and Decision-Making under Uncertainty for Long-Horizon Robot Autonomy

PhD Position: Safe Planning and Decision-Making under Uncertainty for Long-Horizon Robot Autonomy Join TU Delft and the EU-funded OPERA consortium. Push the frontier of general-purpose AI for robotics and tackle the fundamental tension between long-horizon reasoning and real-time safety guarantees. Job description Robots that operate for long periods among people must plan beyond the next few seconds while staying safe. They need to predict how their surroundings and the people in them will evolve, know how reliable these predictions are, and adapt their plans when the predictions fail. In the EU-funded OPERA project, TU Delft contributes to General-Purpose AI for robotics by developing methods that combine fast, reactive behavior with more deliberate reasoning, planning, and decision-making. In this PhD position, you will develop planning and decision-making methods that let robots act safely and reliably over long time horizons when their knowledge of the environment and of the people around them is incomplete. The research sits at the intersection of model-based planning and control and learning-based models, and addresses the trade-off between safety, performance, and real-time computation. You will formulate the methods and analyze their guarantees and failure modes. This research is connected to OPERA's tasks on GPAI-powered planning and decision-making, on long-term reasoning in human-centered environments, and on adaptive human-robot and robot-environment interaction. Depending on your background, the work may draw on optimal control and model predictive control, motion planning, probabilistic modeling and estimation, or learning-based methods. The emphasis is on methods with a clear formulation and well-understood failure modes, validated in simulation and on real robots. You will work in the Cognitive Robotics Department at TU Delft under the supervision of Prof. Robert Babuška and Dr. Laura Ferranti, and you will be embedded in the Reliable Robot Control Lab (https://r2clab.com/). You will collaborate with OPERA partners across Europe on simulation, perception, and the OPERA open-source toolbox. Job requirements The ideal candidate for this PhD position has a strong background in control, optimization, or motion planning and is motivated by the question of how robots can plan safely and reliably over long time horizons. We welcome applicants from all backgrounds who want to work at the intersection of planning, control, and learning. You have: An MSc degree in Systems and Control, Robotics, Applied Mathematics, or a closely related field. A strong interest in safe planning and decision-making for robots, at the intersection of planning, control, and learning. A passion for rigorous, theoretically grounded research, combined with an eagerness to validate ideas in simulation and on real robots. Strong analytical and mathematical skills, in particular in optimization, probability and statistics, and dynamical systems. Excellent communication skills and proficiency in English (written and verbal), as required for academic publication and international collaboration. You are particularly encouraged to apply if you have experience in one or more of the following areas: Hands-on experience implementing optimization-based planners or controllers (e.g., model predictive control with acados or CasADi) in C++ or Python. Safe or robust control and planning under uncertainty. Learned dynamics or prediction models, model-based reinforcement learning. Uncertainty quantification, probabilistic or Bayesian modeling, state estimation. Long-horizon or hierarchical planning, prediction of human or multi-agent behavior. Real-robot deployment with ROS or ROS 2. As part of OPERA, you will travel to meet and collaborate with the project’s European partners and attend regular consortium meetings. We particularly encourage applications from women and other underrepresented groups, as we are committed to building a diverse and inclusive research environment. TU Delft (Delft University of Technology) Working at TU Delft means contributing to solutions that really make a difference. For over 180 years, we have been training engineers who make an impact worldwide in companies, government bodies, or as entrepreneurs. Our alumni turn knowledge into concrete solutions for the challenges of today and tomorrow. These challenges are changing rapidly. That is why we focus on themes such as energy, climate, digitalisation, artificial intelligence (AI), and smart mobility every day. Our education and research are directly aligned with what society needs now and in the future. At TU Delft, our people make the difference. With their knowledge and curiosity, our staff provide a high-quality education and conduct pioneering research that extends beyond the campus. You will have the opportunity to take the initiative, work with others, and grow as a professional. Working at TU Delft means join an international community of professionals and students. Together, we create knowledge, innovations, and solutions that help move the world forward. Faculty Mechanical Engineering From chip to ship. From machine to human being. From idea to solution. Driven by a deep-rooted desire to understand our environment and discover its underlying mechanisms, research and education at the ME faculty focusses on fundamental understanding, design, production including application and product improvement, materials, processes and (mechanical) systems. ME is a dynamic and innovative faculty with high-tech lab facilities and international reach. It’s a large faculty but also versatile, so we can often make unique connections by combining different disciplines. This is reflected in ME’s outstanding, state-of-the-art education, which trains students to become responsible and socially engaged engineers and scientists. We translate our knowledge and insights into solutions to societal issues, contributing to a sustainable society and to the development of prosperity and well-being. That is what unites us in pioneering research, inspiring education and (inter)national cooperation. Click here to go to the website of the Faculty of Mechanical Engineering. Do you want to experience working at our faculty? These videos will introduce you to some of our researchers and their work. Conditions of employment Doctoral candidates will be offered a 4-year period of employment in principle, but in the form of 2 employment contracts. An initial 1,5 year contract with an official go/no go progress assessment within 15 months. Followed by an additional contract for the remaining 2,5 years assuming everything goes well and performance requirements are met. Salary and benefits are in accordance with the Collective Labour Agreement for Dutch Universities, increasing from €3059 - €3881 gross per month, from the first year to the fourth year based on a fulltime contract (38 hours), plus 8% holiday allowance and an end-of-year bonus of 8.3%. As a PhD candidate you will be enrolled in the TU Delft Graduate School. The TU Delft Graduate School provides an inspiring research environment with an excellent team of supervisors, academic staff and a mentor. The Doctoral Education Programme is aimed at developing your transferable, discipline-related and research skills. The TU Delft offers a customisable compensation package, discounts on health insurance, and a monthly work costs contribution. Flexible work schedules can be arranged. Will you need to relocate to the Netherlands for this job? TU Delft is committed to make your move as smooth as possible! The HR unit, Coming to Delft Service, offers information on their website to help you prepare your relocation. In addition, Coming to Delft Service organises events to help you settle in the Netherlands, and expand your (social) network in Delft. A Dual Career Programme is available, to support your accompanying partner with their job search in the Netherlands. Additional information For more information, please reach out to Laura Ferranti and Robert Babuska [l.ferranti@tudelft.nl, r.babuska@tudelft.nl] Please do not use LinkedIn and other channels for applications or inquiries. Application procedure Are you interested in this vacancy? Please apply no later than 15 November 2026 via the application button and upload the following documents: A letter of motivation explaining why you are the right candidate for this project An updated cv Exam transcipts from your Bachelor and Master A copy of your MSc thesis, or draft MSc thesis (or a link to access it) You can address your application to Laura Ferranti and Robert Babuska. Doing a PhD at TU Delft requires English proficiency at a certain level to ensure that the candidate is able to communicate and interact well, participate in English-taught Doctoral Education courses, and write scientific articles and a final thesis. For more details please check the Graduate Schools Admission Requirements. Please note: #EUfunded This is an EU funded project, named OPERA, with project number 101298363, within program HORIZON/HORIZON-CL4-2025-04-DIGITAL-EMERGING-07 You can apply online. We will not process applications sent by email and/or post. As part of knowledge security, TU Delft conducts a risk assessment during the recruitment of personnel. We do this, among other things, to prevent the unwanted transfer of sensitive knowledge and technology. The assessment is based on information provided by the candidates themselves, such as their motivation letter and CV, and takes place at the final stages of the selection process. When the outcome of the assessment is negative, the candidate will be informed. The processing of personal data in the context of the risk assessment is carried out on the legal basis of the GDPR: performing a public task in the public interest. You can find more information about this assessment on our website about knowledge security. Please do not contact us for unsolicited services. Faculty/Department: Faculty of Mechanical Engineering Salary range: €3059 - €3881 Hours per week: 38-40 FTE: 1,0 Submission is possible until: 15 Nov 2026 ID job: 3993

PhD Project Research Report: Safe Planning for Long-Horizon Robot Autonomy at TU Delft

Quick Decision

How can a robot plan beyond reliable short-term predictions of people and surroundings while meeting safety and computation deadlines? This control-and-planning doctorate combines analysis of theoretical guarantees, real-time optimization and real-robot validation under Robert Babuška and Laura Ferranti. It fits control, robotics or applied-mathematics preparation. The practical decision also depends on the formal continuation review, the discrepancy between advertised pay and the dated national table, and the still-unspecified platform and first-year scope. The immediate action is to prepare a mathematically explicit planning/control example and apply online by 15 November 2026, vacancy 3993. The employer gives no closing clock or timezone. The public advert contains Apply; the original read-only visit to that link returned the careers homepage, so authenticated uploads and submission acceptance remain unverified. Official OPERA safe robot planning vacancy 3993

China-trained applicants are not excluded by the published vacancy: foreign-equivalent Master's degrees are accepted and the researcher immigration route covers non-EU doctoral candidates. Admission, English evidence, residence permission and knowledge-security assessment remain individual decisions. No admission probability is inferred. TU Delft doctoral admission and English requirements IND researcher residence permit Directive EU2016/801 Nuffic comparison of Chinese qualifications

Research Problem and Doctoral Work

What is unsolved

A robot can follow an optimized route accurately and still fail if a pedestrian changes intention, a learned prediction is miscalibrated, or optimization returns too late. Extending the horizon compounds those errors. The advertised problem therefore concerns the relationship between safety, performance and computation when environmental knowledge is incomplete; it is not simply an LLM application or a reward-maximization project. The contribution must make its assumptions and failure modes intelligible. The programme connects to EU OPERA, grant 101298363, on general-purpose AI planning, long-term reasoning and adaptive interaction. Official OPERA safe robot planning vacancy 3993

Comparing method choices

The three method families address different questions. Hierarchical MPC separates slower long-horizon planning from fast tracking. Scenario MPC constrains joint trajectory risk relative to a specified prediction distribution. Topology-guided parallel planning preserves distinct passing choices. Compare them at fixed prediction assumptions, horizon, obstacle scope and compute budget, then assess risk, task performance and missed deadlines. These distinctions concern time scales, risk accounting and behavioural alternatives; they are not three required implementations.

Work and plausible outputs

The vacancy advertises formulation and analysis of planning/decision methods, followed by simulation and real-robot validation. Depending on background, methods can include optimal control, MPC, motion planning, probabilistic estimation or learning-based models. European partner collaboration concerns simulation, perception and the OPERA open-source toolbox, and includes travel and consortium meetings. These are confirmed activities; a specific robot, dataset, LLM architecture or publication count is not promised. Official OPERA safe robot planning vacancy 3993

Evidence suggests three possible research angles for discussion with the supervisors: compare calibrated versus shifted human-motion predictions at equal risk budgets; connect a slower planner to a fast safety-aware execution layer and report solver overruns as well as average speed; or preserve several discrete passing strategies before optimizing continuous motion. These are analyst proposals derived from the papers, not assigned OPERA deliverables. Appropriate outputs could be a precisely stated guarantee, failure-case analysis, reproducible planner and measured demonstration. Successful simulation alone would not establish human-interaction safety. Embedded Hierarchical MPC for Autonomous Navigation, author full text Scenario-Based Motion Planning with Bounded Probability of Collision, author preprint Topology-Driven Parallel Trajectory Optimization in Dynamic Environments, 2024 v2 author version

School, Programme and Field Context

TU Delft is the employing university and the doctoral degree-awarding institution. The host is the Faculty of Mechanical Engineering, Department of Cognitive Robotics, within Learning and Autonomous Control. OPERA partners are collaborators; the material does not establish a joint or double doctorate. The ME Graduate School combines independent research and dissertation work with doctoral education, including research, discipline and transferable skills, targeting completion within four years. The current faculty page calls the educational component 45 (ECTS) credits; confirm the applicable courses, recognised activities and units in the registration plan rather than converting this into a taught Master's workload. Mechanical Engineering doctoral programme Cognitive Robotics Learning and Autonomous Control group

QS lists TU Delft 48th overall in the 2027 edition. The university's subject table still explicitly reports QS 2023 Mechanical/Aeronautical/Manufacturing Engineering: 3rd and ARWU 2023 Automation and Control: 33rd. These dated figures support an established engineering environment; they are not a verified 2026 robotics ranking. The more direct subject evidence is the active control group, robot infrastructure and recent methods papers. QS TU Delft university ranking 2027 TU Delft historical subject-ranking table

The lab's programme spans nonlinear MPC, distributed optimization, reinforcement learning and games for reliable coordination around humans. IFR's September 2026 release describes over 600,000 industrial-robot installations in 2025 and an operational stock near five million. This is evidence of a substantial engineering sector, not a count of autonomous-service-robot jobs or a forecast for this PhD. Transferable outcomes include optimization, real-time software, uncertainty evaluation and experimental validation; academic robotics, research institutes and industrial autonomy/control are credible directions. Public data do not establish this lab's placement rate. TU provides career counselling, workshops and events. Reliable Robot Control Lab research programme IFR World Robotics2026 public industry statistics TU Delft doctoral career perspectives

Supervisor and Lab Assessment

Ferranti leads the Reliable Robot Control Lab and is listed as an associate professor; her department highlights cooperative control, real-time optimization, fault tolerance and mobile-robot planning. Babuška's group profile highlights reinforcement learning, nonlinear control and data-driven models. The pairing connects formal control with learning, but neither profile establishes meeting frequency, current supervision load or the division of daily mentoring. Cognitive Robotics Learning and Autonomous Control group Laura Ferranti at Reliable Robot Control Lab Laura Ferranti TU Delft research output

Three recent contributions and their limits

Embedded Hierarchical MPC for Autonomous Navigation (2025). Benders and colleagues, including both advertised supervisors, co-design a slow long-horizon planner and fast tracker, aligning constraints and reference updates. The paper reports a fivefold planning-horizon increase against its single-layer comparator and embedded quadrotor experiments. Its theoretical feasibility and tracking guarantees assume no model mismatch; the tested environment uses static maps and motion capture. Robustness observed experimentally is not a proof for uncertain human motion. The project connection is the engineering and theoretical interface between slower reasoning and fast execution. Reading covers selected formulation/method/results passages and the conclusion of the 9 May 2025 v4 author manuscript, without proof reproduction. Fivefold refers to horizon; the reported travel-time comparison also uses different start/goal positions and is not a general speed-up factor. Embedded Hierarchical MPC for Autonomous Navigation, author full text

Scenario-based motion planning with bounded probability of collision (IJRR 2025). De Groot, Ferranti and colleagues constrain the joint collision probability across the horizon and obstacles using sampled uncertainty, rather than treating each time step's risk independently. This makes risk-budget comparisons more meaningful. The reviewed 2023 author version assumes an accurate uncertainty distribution, can include predicted interactions in the joint distribution but not a probability measure that directly depends on planning decision variables, and notes heavy sampling cost at very small risk levels. Its reported demonstrations are simulations. The 2025 journal record confirms publication, but the final PDF was inaccessible; details are not silently attributed to the final revision. Scenario-Based Motion Planning with Bounded Probability of Collision, author preprint 2025 IJRR bibliographic record for Safe Horizon MPC

Topology-driven parallel trajectory optimization (4 October 2024 v2 author version). De Groot, Ferranti and colleagues preserve distinct passing behaviours through homotopy guidance and constrained parallel local planners, reducing indecision. Real-robot experiments involved five pedestrians. The discussion reports remaining simulation collisions under prediction mismatch and leaves explicit interaction and social norms unresolved. Thus the contribution supports planning alternatives and consistency, not universal safety. Coauthorship is verified; the papers do not establish each supervisor's individual experimental role. Topology-Driven Parallel Trajectory Optimization in Dynamic Environments, 2024 v2 author version

The lab lists DART, a HoverGames drone, an autonomous car and ground robots. Those resources strengthen feasibility, but do not establish which are allocated to this vacancy. Access, technical support, compute budget and the interface to partners' perception systems need confirmation. Reliable Robot Control Lab robot platforms

Funding, Application, and Living Cost

The appointment is four years in principle: an initial 1.5-year contract, a formal go/no-go within 15 months, and a further 2.5 years conditional on satisfactory progress. The vacancy quotes €3,059–€3,881 gross/month at 38 hours, with 8% holiday allowance and 8.3% year-end bonus; its summary separately states 38–40 hours. The official salary table effective 1 July 2026 instead lists doctoral scale P steps €3,204, €3,706, €3,873 and €4,051. The report preserves this discrepancy: the higher table is a verification reference, not a substituted offer. Net pay, pension, tax treatment and future increases are not calculated. Official OPERA safe robot planning vacancy 3993 Dutch university salary scales effective 1 July 2026

This is an advertised employee position rather than an external scholarship. The generic university fees page nevertheless discusses a €11,000 programme contribution and variable annual bench fees, requiring agreement on calculation, amounts and payer at programme start, preferably before the first working day, with faculty waivers possible; bench fees go to the institute, or to the candidate if not institute-funded. The vacancy does not specify the fee allocation. Written confirmation that the department/project covers any applicable doctoral and laboratory charges is preferable to assuming that a salary implies every fee is waived. Taught non-EU Master's tuition is irrelevant to this employment comparison. TU Delft PhD fees and funding routes

The international-employee housing page quotes July 2026 single studios at €785–€980/month, including utilities, internet and municipal taxes, for initial accommodation up to six months, subject to availability. It is not a four-year rental guarantee, does not permit children/pets in those units and requires a €285 fee, normally non-refundable except when TU Delft has not uploaded the details to ROOM. The service is for people arriving directly from abroad, lasts two weeks to six months with no extension, and uploads occur no earlier than seven weeks before the rental start, subject to availability. The rental start cannot be changed after booking. These restrictions concern this housing service, not a general ban on family relocation. Build the budget from a real lease, adding food, insurance, transport and personal costs, with separate cash for deposits and relocation. No total living-cost or net-savings estimate is justified by unspecified assumptions. The national student benchmark of €1,000–€1,500 is older, based partly on Nibud 2024, and can understate an incoming employee's housing needs. Health insurance is mandatory. Deposits, travel, immigration costs and housing after month six require cash reserves. TU Delft international employee short-stay housing TU Delft international employee insurance guidance Nuffic Study in NL cost reference

Applicant Fit and Preparation

Required: a relevant MSc in Systems and Control, Robotics, Applied Mathematics or a close field; strong optimization/probability/dynamical-systems foundations; English communication; motivation for theory and real experiments. TU's general English threshold is IELTS Academic 7.0 overall, 6.5 each; TOEFL taken from 21 January 2026 uses 5 overall, 4.5 each, while earlier tests require 100/22. Results must be within the two-year validity window at appointment; English-taught MSc holders are in principle exempt, with possible local assessment. Results are requested later in selection, not with the initial application. Cambridge C1 Advanced or C2 Proficiency at 185 is also accepted. TOEFL scores must come from one test, with no MyBest; IELTS One Skill Retake is accepted, but Online and Indicator are not. Other testing organisations are not accepted. Native speakers are exempt; if an English-medium MSc holder is asked for a placement test, it can only be taken at TU Delft, not online. Nuffic's China table lists Master's certificates as potentially HBO or WO Master's level, so the actual degree must be evaluated. Official OPERA safe robot planning vacancy 3993 TU Delft doctoral admission and English requirements Nuffic comparison of Chinese qualifications

Preferred, not cumulative requirements: acados/CasADi optimization in C++/Python, robust planning/control, learned dynamics or model-based RL, Bayesian estimation, hierarchical or human-behaviour planning, and ROS/ROS2 real-robot work. Suggested preparation: present one reproducible controller with explicit constraints, timing distributions and failure cases; explain a paper's uncertainty assumptions and a modest extension. Purely high-level AI demos without control mathematics, or proof-only work without interest in deployment, provide weaker evidence for the advertised combination. This is a background comparison, not an assessment of a particular applicant. Official OPERA safe robot planning vacancy 3993

Apply online with a motivation letter, updated CV, BSc/MSc transcripts and MSc thesis or draft/link. The advert does not request a separate research proposal; the suggested research angles support preparation rather than add a mandatory document. Questions go to l.ferranti@tudelft.nl and r.babuska@tudelft.nl; the advert rejects applications by email/post and requests no LinkedIn inquiries. TU's knowledge-security review considers the research field and study/work affiliations in the final selection stage. Non-EU residence permission requires the employer's approved research, suitable qualification and contract/hosting arrangements; it is not automatic. Official OPERA safe robot planning vacancy 3993 IND researcher residence permit Directive EU2016/801 TU Delft knowledge security and recruitment assessment

Next Actions and Questions to Verify

Before applying

Submit the project-specific materials by 15 November 2026, after checking the live vacancy. If relevant, ask whether the foreign MSc and English-medium exemption meet registration requirements. Do not wait for a speculative scholarship or use a generic AI proposal.

At interview

• Which robot and human-motion/prediction interface will the first-year work use, and what is already operational?
• Which guarantee is sought under model or distribution shift, and what benchmark/failure metrics define progress?
• How are day-to-day mentoring, meetings and responsibilities divided between Ferranti, Babuška and OPERA partners?
• What are the travel, compute and experimental-support budgets, and which partner visits are required?

Before accepting an offer

• Obtain the current P-scale salary, FTE/workweek, fee payer and complete 1.5/2.5-year contract terms in writing.
• Clarify the 15-month continuation criteria and what happens after delays, leave or a no-go decision.
• Confirm immigration documentation, relevant security assessment, actual housing offer and a plan after six months.

Evidence and Sources

The original assessment is dated 25 October 2026 UTC; the authenticated recruitment was collected on 24 October. This edit reuses the saved reports and records without a fresh vacancy check or new paper reading. The three discussions retain their version-specific selected-section scopes, including an inaccessible final IJRR PDF and unreproduced proofs. The earlier application-link visit did not verify authenticated uploads or submission acceptance.

Project, training and institution

• Official OPERA safe robot planning vacancy 3993
• TU Delft doctoral admission and English requirements
• TU Delft PhD fees and funding routes
• Mechanical Engineering doctoral programme
• QS TU Delft university ranking 2027
• TU Delft historical subject-ranking table
• Cognitive Robotics Learning and Autonomous Control group
• TU Delft knowledge security and recruitment assessment

Supervisors, lab and research

• Laura Ferranti at Reliable Robot Control Lab
• Reliable Robot Control Lab research programme
• Reliable Robot Control Lab robot platforms
• Laura Ferranti TU Delft research output
• Embedded Hierarchical MPC for Autonomous Navigation, author full text
• Scenario-Based Motion Planning with Bounded Probability of Collision, author preprint
• Topology-Driven Parallel Trajectory Optimization in Dynamic Environments, 2024 v2 author version
• 2025 IJRR bibliographic record for Safe Horizon MPC

Funding, mobility, living and careers

• Dutch university salary scales effective 1 July 2026
• IND researcher residence permit Directive EU2016/801
• Nuffic comparison of Chinese qualifications
• TU Delft international employee short-stay housing
• TU Delft international employee insurance guidance
• Nuffic Study in NL cost reference
• IFR World Robotics2026 public industry statistics
• TU Delft doctoral career perspectives

Known ordered bibliographies from the saved research record

• Dennis Benders; Johannes Köhler; Thijs Niesten; Robert Babuška; Javier Alonso-Mora; Laura Ferranti. Embedded Hierarchical MPC for Autonomous Navigation. IEEE Transactions on Robotics 41, 3556–3574 (2025); assessed author manuscript v4 dated 9 May 2025. Cited version.
• Oscar de Groot; Laura Ferranti; Dariu M. Gavrila; Javier Alonso-Mora. Scenario-based motion planning with bounded probability of collision. The International Journal of Robotics Research 44(9), 1507–1525 (2025); assessed 2023 author preprint, final PDF not read. Cited version.
• Oscar de Groot; Laura Ferranti; Dariu M. Gavrila; Javier Alonso-Mora. Topology-Driven Parallel Trajectory Optimization in Dynamic Environments. IEEE Transactions on Robotics, accepted author version v2 dated 4 October 2024; no final issue details inferred. Cited version.