PhD in Urban Robot Navigation and Mapping: Geospatial World Models

PhD in Urban Robot Navigation and Mapping: Geospatial World Models

Quick Decision

Promising with material uncertainties. This is a distinctive robotics and geospatial AI project: it proposes that urban robots use existing map context and live sensor observations together, continuously improve semantic maps, and navigate unfamiliar places with less dependence on costly manual mapping. It names real-world applications in delivery, infrastructure assessment and city monitoring and lists four Cardiff contacts whose profiles span embodied AI, robot planning, formal reasoning and human-centred computing. The strongest match is a quantitatively trained computer science, robotics, engineering or related applicant who can demonstrate programming, perception/mapping and experimental work.

The advertised route is a competitive EPSRC Doctoral Landscape Award (DLA) for an October 2027 start, with a stated deadline of 5 January 2027. Cardiff’s scheme page says the final DLA cohort starts in 2027/28, offers nine studentships per year across the university and structures each award over 3.5 years. That confirms a scheme-level route, not an award for this specific applicant or a quota for this topic. The project’s 2027/28 stipend is unpublished. Cardiff’s general UKRI policy says successful international awardees receive a fee discount covering the home/international difference and that broadly up to 30% of each DTE intake may be international; neither statement confirms that this project has an international slot. Ask the named contact and the PGR funding team to confirm award terms, fee coverage and international eligibility in writing before relying on them. EPSRC DLA · Cardiff UKRI studentships · UKRI student support

Research Problem and Doctoral Work

The project addresses a practical limitation of autonomous urban robots: deployment often depends on having a detailed, prebuilt map, which is costly to create and maintain as streets and infrastructure change. The proposed system would bring together a robot’s live multimodal observations and large-area geospatial context, including OpenStreetMap (OSM) features such as pedestrian paths, crossings, road networks and land use. The robot is framed as both a consumer of map information and a source of updates for navigation and city-facing uses.

The advertised work has four connected parts:

1. Continual urban mapping: generate and update persistent, semantically rich, OSM-compatible representations from routine robot operation, including methods for detecting and incorporating environmental changes.
2. Geospatially grounded world models: fuse onboard sensors with geospatial prior data and represent urban geometry, semantics, topology and uncertainty for decision-making.
3. Map-light navigation: use the representation to predict traversability, identify useful navigation affordances, reason about likely outcomes and adapt to previously unseen environments without a manual mapping campaign.
4. Real-world evaluation: test the value of robots for applications such as delivery, infrastructure assessment and city-scale monitoring. The notice names Robot.com as an industry partner but does not specify the partner’s data, deployment, supervision, intellectual-property or funding commitments.

A useful scientific framing is the coupling between prior knowledge, live sensing and safe action. OSM features can be incomplete, stale or geometrically imprecise; onboard observations can be noisy or occluded; localization error can misalign the two. A world model must therefore say what a robot believes, how uncertain it is, when it trusts or revises a prior, and how that uncertainty changes a navigation decision. The work also needs a clear operational definition of “world model”: for example, whether the central output is a predictive latent state, a topological-semantic map, an uncertainty-aware scene graph, or an integrated representation supporting particular tasks. These are research questions to resolve with the supervisors, not methods guaranteed by the notice.

Current research points to demanding evaluation questions. A 2025 survey of embodied-AI world models identifies gaps in unified datasets, physically meaningful evaluation, real-time efficiency and long-horizon temporal consistency/error accumulation. A 2024 preprint demonstrates a two-stage persistent semantic mapping approach in two indoor environments, with repeated-visit object updates; it is a nearby design pattern, not evidence of city-scale transfer. World-model survey · Persistent semantic mapping study

Promising research angles include: (a) quantifying when and how a robot should defer to geospatial priors versus newer sensor evidence; (b) maintaining temporally consistent semantic/topological maps when localization, object detection and map matching are uncertain; and (c) comparing navigation with no prior map, a static prior, and an incrementally corrected prior across new areas and changing conditions. The project notice does not prescribe these exact questions. The first year would benefit from selecting one primary use case, a tractable platform and a small set of measurable outcomes—such as map-update precision/recall, localization or traversability error, navigation success/safety, map freshness, and generalization to held-out streets—before attempting multiple city applications.

Because OSM data is distributed under the Open Database License, publicly sharing derived works can require attribution and may have database-related conditions. The intended release of maps, data and code should be planned early with licensing and provenance in mind. OSM attribution guidance

School, Programme and Field Context

Cardiff University is the degree-awarding institution and is located in Cardiff, Wales, UK. The opportunity notice and the Computer Science and Informatics PGR page use that school/programme name; some current staff profile role metadata also displays “School of Computational and Mathematical Sciences.” Confirm the project’s administrative home for the October 2027 intake. Cardiff location · Computer Science and Informatics PhD

The PGR programme lists a standard full-time PhD duration of three years and part-time duration of five years, with four entry points each year. It describes seminars, annual skills training, progress checks every six months and an annual review. Its advertised-studentship deadlines can replace standard entry deadlines. Cardiff’s DLA page describes financial support structured over 3.5 years, so ask how the registration period, award period, fee liability and any writing-up period align. The programme page says funded studentships normally cover a stipend and tuition at the home/EU rate; the DLA page confirms scheme-level fully funded support but does not publish this project’s actual award letter or stipend.

For institutional context, Cardiff reports that 96% of its Computer Science and Informatics REF 2021 submission was judged world-leading or internationally excellent. The university’s current summary lists QS World University Rankings =179 overall for 2027 and ARWU 201–300 overall for 2026; THE places Computer Science 201–250 for 2026. These are broad, differently scoped measures. The REF result is school-level; none directly ranks this project or its supervisors. Cardiff REF 2021 result · Cardiff rankings · THE Computer Science ranking

Supervisor and Lab Assessment

The advert names Victor Romero-Cano, Nico Potyka, Juan Hernandez Vega and Nervo Verdezoto Dias. It does not clearly allocate lead and co-supervisor roles, so treat the following as research fit rather than a confirmed supervisory structure.

• Victor Romero-Cano is the closest match to the project’s core. His Cardiff profile identifies interests in world representations and models, spatial and embodied AI, robot perception and decision-making, sensor fusion and autonomous systems. He co-leads the Cardiff University Computational Robotics group and manages the Human-Centred Computing and Robotics laboratory. Profile
• Juan Hernandez Vega works on motion planning, safe navigation, social navigation, field robotics and robot learning. His prior mapping/planning research is directly relevant to map-building under uncertainty and navigating without a complete prior map. Profile
• Nico Potyka works in formal and probabilistic reasoning, knowledge representation, explainable AI and robustness. This could complement uncertainty-aware decisions, but his public profile is not primarily a robotics profile. Profile
• Nervo Verdezoto Dias leads the Human-Centred Computing unit and works in HCI, participatory design, socio-technical evaluation and human-robot interaction. This expertise could help if planners, local authorities, delivery operators or residents participate in defining map usefulness and impact; no such study is specified in the advert. Profile

Recent papers provide more concrete evidence than profile keywords alone:

Paper and contribution	Relevance and limit
Romero Cano and co-authors, “Barrios 4D” (2025), derive settlement morphology measures from mobile LiDAR, including street-canyon metrics, and analyse a city case.	Connects sensing to urban information needs and planning; it does not demonstrate autonomous robot mapping or OSM integration. Paper
Romero Cano and co-authors, probabilistic camera–LiDAR depth fusion (2023), fuse raw streams through a conditional random field with geometry and appearance terms and evaluate dense depth on KITTI.	Supports multimodal perception and uncertainty experience; the work was completed in a prior institutional context and is not a persistent mapping system. Paper
Hernandez Vega, Verdezoto and colleagues, online social navigation (2023), combine a social heatmap/world model, multilayer sampling-based planning and path following, with simulation benchmarks and Pepper robot trials.	Shows integration of a representation and planner on a physical platform; its target is indoor crowded human spaces rather than geospatial urban navigation. Paper
Hernandez Vega and co-authors, uncertainty-aware online mapping and motion planning (2022), incrementally build an uncertainty-aware map and replan kinodynamically feasible, probabilistically safe paths in belief space.	Closely informs the map–uncertainty–planning interface, but does not solve persistent city-scale updates or OSM prior fusion. Paper

The group and centre environment appears active in computational robotics, embodied AI, human-centred systems and autonomy. Cardiff’s Computational Robotics group describes real-robot evaluation and current 3D-world/embodied-AI work; IROHMS spans robotics, AI and human-machine systems. Cardiff’s Autonomous Systems and Robotics Laboratory lists five TurtleBot3 mobile platforms with LiDAR and camera, KUKA platforms, drones and a USV. That laboratory is housed in Engineering, so it is evidence of nearby institutional capability, not a guarantee that this Computer Science project will use those robots. Ask which platform and equipment budget are committed. Computational Robotics group · IROHMS research · Autonomous Systems and Robotics Laboratory

Funding, Application, and Living Cost

The notice labels the studentship competitive and tells applicants how to identify the 2026 EPSRC DLA route in the online form. Cardiff’s official page confirms Computer Science is within the DLA remit, October 2027 is the final 2027/28 cohort, support is structured over 3.5 years, and nine studentships are available annually across the scheme. Schools set their own project processes. The available information does not reveal how many of those awards are assigned to this project, the selection rate, or whether an applicant meeting minimum criteria will receive funding. EPSRC DLA

Stipend and fees. UKRI’s current minimum full-time stipend is £21,805 from October 2026, with a £5,238 home-fee contribution for 2026/27. These are current baselines only: the exact 2027/28 rate and this project’s stipend are not yet published. UKRI contribution is capped at the home-fee rate. Cardiff’s general UKRI international page says the university covers the home/international fee gap for successful international awardees and that broadly up to 30% of each DTE intake may be international. UKRI guidance says international limits vary by research council. The policy is encouraging but does not establish that the specific EPSRC DLA project will recruit an international candidate or that this applicant will be one of the capped awardees. Obtain written confirmation of the tuition amount charged to the student, any fee discount, eligibility category and stipend before accepting. UKRI student support · Cardiff UKRI studentships · UKRI training-grant guidance

Living costs. Cardiff’s 2026/27 calculator estimates postgraduate costs at £1,402.98 per month/£16,835.78 per year for university halls over a 51-week contract, or £1,125.91 per month/£13,510.92 per year for a private shared house over 52 weeks. The private estimate is based on a four-bedroom shared house and costs checked in May 2026. Travel is excluded, and the university warns that estimates vary. Against the current £21,805 stipend baseline, the halls estimate leaves about £4,969 before travel and other unbudgeted expenses. Use this only as a planning scenario: the project starts in 2027/28, when both costs and UKRI rates may differ. Cardiff living costs

Application. The stated deadline is 5 January 2027, with an October 2027 start. The notice does not give a deadline hour or time zone; because the application is to Cardiff, ask the school whether it means local UK time and what cutoff applies. The notice asks for an online Computer Science application, a personal statement no longer than 500 words, CV, degree certificates/transcripts with English translations if needed, two references and proof of English where applicable. It asks applicants to answer “No” to the self-funding question, name the 2026 EPSRC DLA funding route, specify the project title and supervisors, and paste the project description in the designated research-proposal field. Cardiff’s generic programme page describes research proposals of up to 2,000 words; verify whether the project-specific form requires anything beyond the pasted description. The opportunity notice’s 500-word statement prompt asks about research ambition, preparation for doctoral study and why this project matters. Submit early enough to resolve document and referee issues.

Applicant Fit and Preparation

Formal criteria. The opportunity notice asks for a 2:1 honours degree or master’s in computing or a subject with a significant quantitative component; it gives mathematics, engineering, economics and psychology as examples and says appropriate professional experience may be considered. The PGR programme independently describes a 2:1 or master’s in computing or a related field and requires IELTS 6.5 overall with at least 6.0 in each skill for applicants to whom that English standard applies. Some research areas may require degree-level mathematics. Confirm qualification equivalence and accepted English evidence with admissions. Computer Science and Informatics PhD

Preferred evidence, inferred from the research—not advertised prerequisites. A strong application would show a working grasp of software implementation and evaluation, quantitative reasoning, robotics or autonomous systems, and at least one of computer vision, LiDAR, SLAM/localization, GIS/geospatial data, machine learning or planning. Useful preparation includes coordinate transforms and 3D geometry, probabilistic estimation, graph/topological representations, Linux/Git, robot middleware and reproducible experiments. The notice does not require a particular programming language, robot framework or prior contribution to OSM, so do not present these suggestions as formal eligibility rules.

A focused portfolio or statement example could describe a sensor-fusion, map-update or navigation project with a clear baseline, quantitative evaluation, failure cases and limitations. Applicants from GIS, urban analytics or data science should make technical implementation and robotics readiness visible; applicants from robotics or computer science should show awareness of geospatial provenance, map quality and the people who may use urban outputs. In the 500-word statement, answer all three prompts directly and prioritize specific evidence over a generic account of interest. At interview, prepare a realistic first-year prototype and explain how it would detect stale map priors, quantify uncertainty and evaluate safe generalization.

Field and Career Outlook

The opportunity sits at an intersection of mobile robotics, embodied AI, geospatial computing, semantic mapping and urban technology. The field is active, but world-model terminology covers different representations and objectives; the thesis needs a testable definition tied to navigation or mapping performance. Long-horizon consistency, real-time compute, robust perception and fair out-of-distribution evaluation are plausible research bottlenecks. A nearby 2024 indoor mapping preprint demonstrates a map-once/update-semantics pattern on a Fetch robot, but its bounded indoor sites underline why the advertised city-scale claim needs staged validation rather than assumed transfer. World-model survey · Persistent semantic mapping study

The UK Government’s 2025 AI labour-market survey reports persistent technical skills gaps and difficulty filling some AI roles. The International Federation of Robotics reports growth in global professional service robot shipments in 2025, particularly transportation and logistics. These sources support broad relevance for AI and service-robot capabilities; they do not predict robotics hiring in Cardiff, guarantee a role, or show that a city delivery market will adopt this project’s methods. Plausible career directions include robotics/autonomy R&D, perception and SLAM, AI/ML engineering, geospatial or urban technology, applied research and academic work. The strongest transferable evidence would be a reliable deployed prototype, sound evaluation, clear research outputs and collaboration across technical and urban stakeholders. AI Labour Market Survey · IFR service-robot report

Next Actions and Questions to Verify

Before Applying

Email Dr Victor Romero-Cano with a short summary of relevant technical experience and ask:

1. Is this project definitely attached to the 2027/28 final EPSRC DLA cohort, and what is the selection process and number of awards available to this project?
2. What is the expected stipend for an October 2027 starter, the length of funding, home-fee coverage and any separate research/training allowance?
3. Are international applicants eligible for this exact project? Does the DLA have an international award slot for the cohort, and will Cardiff’s fee-gap discount apply to this award?
4. What are the precise application cutoff time and time zone on 5 January 2027? Which form fields require the supplied description, and is an additional proposal required?
5. Who will be lead supervisor and how are the four named contacts expected to contribute?
6. What are the initial robot, sensors, urban test site, data sources and year-one deliverables? How will Robot.com participate, and are there placement, access, confidentiality or IP conditions?

At Interview

Ask how the team will narrow the four workstreams to a tractable thesis; what baselines and safety metrics distinguish useful map-light navigation from a conventional map-based system; how prior-map errors and localization uncertainty are represented; and what real-world access and technical support are guaranteed. Discuss how map updates and derived data will be validated with planners, operators or other end users if those stakeholders are in scope.

Before Accepting an Offer

Review the written award terms for stipend amount and annual progression, fee liability for the applicant’s status, funding end date, registration/writing-up arrangement, research/training costs, international eligibility, and any visa/health surcharge or relocation costs. Confirm the project’s administrative school, supervisor roles and partner conditions. Do not rely on general DLA policy text as a substitute for the individual award offer.

Evidence and Sources

The opportunity notice describes the project tasks and examples. Before applying, confirm project ownership, technical scope and the industry partner's role with the named Cardiff contacts. The sources below support the institutional, funding, supervisor, literature, cost and field context:

• EPSRC Algorithmic Grant Doctoral Landscape Award
• Computer Science and Informatics PhD/MPhil programme
• UKRI support for funded students
• Cardiff UKRI international studentships
• UKRI training-grant holder guidance
• Cardiff living-cost calculator
• Cardiff University rankings
• THE Cardiff subject rankings
• Cardiff Computer Science REF 2021 result
• Cardiff University location
• Victor Romero-Cano profile
• Juan Hernandez Vega profile
• Nico Potyka profile
• Nervo Verdezoto Dias profile
• IROHMS research environment
• Human-centred Computing research area
• Autonomous Systems and Robotics Laboratory
• Cardiff University Computational Robotics group
• Barrios 4D paper record
• Probabilistic camera–LiDAR depth paper record
• Online social robot navigation paper record
• Online mapping and motion-planning paper
• World-model survey
• Persistent semantic mapping study
• OpenStreetMap attribution guidelines
• UK AI Labour Market Survey 2025
• International Federation of Robotics, 2025 professional service robot shipments