Bath Engineering & Design PhD Studentships: Multimodal Spatial Intelligence and Traffic World Models

Bath Engineering & Design PhD Studentships: Multimodal Spatial Intelligence and Traffic World Models

Quick Decision

How can separate roadside views become a coherent account of traffic, and when can that account explain or predict events? This Bath project starts with distributed cameras and sensing, proposing dynamic spatial representations of vehicles, interactions and changing traffic. Its research value is the gap between reliable perception and a useful traffic world model. Tracking a vehicle across views does not by itself demonstrate causal explanation or accurate future prediction.

The opportunity is Strong but verify fit for an applicant with a relevant quantitative degree, competent programming and machine learning, and an interest in rigorous computer-vision evaluation. The formal registration is Mechanical Engineering PhD, with Dr Nic Zhang as the named contact, qz254@bath.ac.uk. The project notice gives 29 November 2026 without a clock; the saved Engineering URSA terms add 23:59 UK time, GMT. Expected entry is 4 October 2027, with alternatives by agreement. Non-UK applicants not currently studying in the UK must meet English requirements before the deadline, without exceptions. The saved materials do not establish whether applications are still being accepted. Engineering URSA terms

Home and Overseas tuition support is available through competition, with an unspecified international cap. A personal award, 2027 maintenance rate and funded months are not confirmed. A separate university page for this exact topic was not identified: official programme, supervisor and funding sources verify their own information. The most useful first discussion concerns the initial research task, real data access, evaluation design and the written award terms.

Research Problem and Doctoral Work

From Multi-View Observations to a Traffic Model

The proposal seeks to fuse observations from distributed cameras and other sensing into a changing spatial representation. Potential methods include tracking, vehicle re-identification, geospatial mapping, three-dimensional understanding, multimodal vision-language models, anomaly detection, explainable AI and prediction. These are possible components, not a promise that every method will be required or delivered. No particular city, private camera feed, sensor agreement or operational deployment has been assigned publicly.

A coherent representation needs to connect observations across place and time while preserving uncertainty. A visually plausible explanation can still be unsupported by the observations; an identity association can be correct without explaining why congestion occurred. A forecast must be evaluated against future observations rather than judged from the quality of a descriptive scene summary. This separation gives the topic a defensible doctoral question, but also makes an overly broad first-year plan risky.

A sensible initial discussion would distinguish a bounded perception problem from the subsequent modelling ambition. For example, improved association under occlusion could support more reliable trajectories, while a later question concerns what those trajectories justify about interactions. These are analytical options, not advertised compulsory work packages. Ask what baseline, outcome and held-out setting would establish a contribution before adding a large multimodal model. Bath's spatial-intelligence themes provide relevant context for sensing, synthesis and world models. Centre research themes

Evaluation, Data and Transferable Value

Useful evaluation goes beyond one identity score. Accuracy, false associations, missed vehicles, temporal consistency, uncertainty and computational cost answer different questions. A traffic forecast additionally needs a defined horizon, relevant comparator and out-of-sample test; explanatory claims need evidence beyond correlations in recorded trajectories. Neither a benchmark leaderboard nor an appealing generated explanation establishes operational safety or causal understanding.

The decisive resource question is which data the student may actually use. Camera coverage, timing, weather, annotation quality, permitted reuse and privacy controls shape the research. Published datasets are evidence of team experience, not a guarantee of a new city's feed or a licensed industrial dataset. The available computing, storage, security arrangements and support for annotation remain unconfirmed. Publication access, code release and dataset licences also need separate answers.

The likely transferable skills include computer vision, scientific software, reproducible evaluation, spatial reasoning and responsible handling of observational data. They are relevant to transport and broader perception research, but specific graduate destinations, salaries and placements were not verified. A successful thesis could make a bounded methodological contribution without delivering a city-wide deployed system. The departmental computational and transport environment supports that possibility; it does not guarantee an industrial contract, energy saving or commercial outcome. Mechanical research environment

School, Programme and Field Context

The University of Bath awards the PhD, through Mechanical Engineering in the Faculty of Engineering and Design, at Claverton Down. The AI theme and cross-disciplinary centre do not change the degree to Computer Science. The full-time programme says most students complete in four years, with a two-year minimum and normally four-year maximum. This academic period is distinct from the URSA funding maximum. Registration usually begins on probation, followed by written work and oral confirmation, supervised research, thesis submission and viva. A studentship is not an employment contract. Mechanical Engineering programme

Nic Zhang's current affiliations include IAAPS, Made Smarter People-Led Digitalisation and the Centre for Spatial Intelligence/RCSI. The centre staff page explicitly places him in Mechanical Engineering. The centre's website being hosted within a different faculty does not establish a different registration. Its directors, Michael Yang and Jason Alexander, are not thereby appointed co-supervisors for this project. Shared centre interests likewise do not reserve equipment or a computing allocation. Current institutional profile, Centre staff

Bath ranks 125th in QS World University Rankings 2027, compared with 132 in 2026. This is an overall institutional signal, not a current ranking of this group, computer vision or Mechanical Engineering. Bath also reports 92% of its research as world-leading or internationally excellent in REF 2021; that is institution-wide. Relevant subject comparisons, supervision completion rates and project graduate outcomes were not established. The group's actual research fit matters more than transferring a general university number to an individual doctorate. QS institutional ranking, Institutional doctoral context

MyDoctoralDevelopment offers research and transferable-skills workshops, communication support and development activities. These opportunities do not specify this student's personal curriculum. Ask how advanced ML, spatial modelling, data ethics, scientific software and confirmation requirements will be organised, and who provides regular guidance in each area. Cross-department collaboration can broaden training without creating a joint degree, guaranteed placement or automatic access to another group's grants. Doctoral development

Supervisor and Lab Assessment

Current Role and Supervision Boundary

Dr Nic Zhang is a Lecturer in Mechanical Engineering, confirmed by the current institutional profile and centre staff record. His published work offers a coherent foundation in multi-camera tracking and transport perception. Historical work with JLR is professional background, not proof of a current industrial sponsor. Likewise, the CHAMP2 record ending 31 March 2025 does not fund this applicant's 2027 doctorate.

The exact co-supervisory team, meeting frequency, student outcomes and responsibility for data or computing support are not stated. Research collaborators and centre directors should not be presented as appointed supervisors. A first email to qz254@bath.ac.uk should show relevant coding/ML experience and ask which part of spatial intelligence will form the first bounded thesis question, rather than assume every advertised possibility is a funded work package.

Three recent original contributions provide complementary evidence:

Camera-link study: cross-view association

The camera-link study is an original ICPAMI 2024 conference chapter published 27 March 2025, DOI 10.1007/978-981-96-3349-4_9. It combines detection, features and single-camera tracks with learned camera links and spatial-temporal constraints. The selected accepted-author Methods 3.1–3.4, Results 4.1–4.5 and Conclusion 5 were read. On the CityFlowV2 six-camera test it reports IDF1 0.6107, while acknowledging a gap to human-annotated state-of-the-art constraints, imperfect zoning, false positives and fragmentation. The authors say hyperparameters were chosen experimentally for best IDF1 and then report test-set results; the separation of tuning, validation and final testing requires further verification. A future thesis should keep these stages separate. The accepted manuscript is not the publisher's final version; institutional and publisher author listings differ, so a definitive full author order is not asserted. This supports tracking expertise, not a proven predictive world model. Camera-link original study

CityFlow ablation: shared data and pipeline

The CityFlow ablation study, an ICMV 2024/SPIE conference paper published 24 February 2025, DOI 10.1117/12.3055212, examines detection, feature extraction, single-camera and multi-camera tracking modules, identity performance and computational demands. The complete institutional original abstract and metadata were read; full methods, results and code were not. Its abstract supports the relevance of automatic spatial-temporal constraints, but does not justify invented numerical gains. It shares CityFlow and pipeline lineage with the preceding study and is not an independent replication in a new city. Ablation original abstract

RoundaboutHD: a single-scene preprint

RoundaboutHD, arXiv 2507.08729v2, 21 July 2025, is an original author preprint, with no verified final peer-reviewed acceptance established here. Selected Methods 3, Results 4 and Conclusion 5 were read. Four cameras record the same ten minutes simultaneously: 40 aggregate camera-minutes, with 512 unique vehicles in one Carmel, Indiana roundabout. Its curved paths and occlusions test association beyond simpler scenes. Re-identification results near 99% are not end-to-end multi-camera tracking accuracy; the ELECTRICITY baseline reports IDF1 28.14, versus 46.16 on CityFlow, an uncontrolled cross-dataset comparison. Single-scene coverage limits generalisation. Training/test identity construction needs code inspection before any leakage claim; parked vehicles are excluded from the tracking comparison. Paper licensing and authors' privacy statements do not independently establish video licensing or a privacy audit. RoundaboutHD v2 methods and results

These readings comprise two original conference papers and a preprint, with disclosed partial reading. They are not three complete journal articles, independent replications or reproduced code. They suggest a productive supervision discussion about evaluation, dataset assumptions and the distance from tracking to explanation, rather than a basis for guaranteeing predictive performance.

Funding, Application and Living Cost

Award, Deadline and Submission

The Engineering URSA round documented in October 2026 is for October 2027, closing 29 November 2026 at 23:59 UK/GMT. The exact project gives the same date without a time; the clock comes from the funding scheme. Expected start is 4 October 2027; another date needs supervisor and Doctoral College agreement, with award retention checked separately. A general international admission deadline three months before entry does not override this scholarship deadline.

Apply through Mechanical Engineering PhD and select University of Bath URSA. Enter the advertised title exactly as “Bath Engineering & Design Studentships: Multimodal Spatial Intelligence and Traffic World Models” and Nic Zhang in the first-choice fields. Applicants may choose at most two projects. Use the exact second-choice project and supervisor fields for another choice. The programme requires two references, at least one academic from the most recent place of study, and a personal statement addressing research motivation, experience and career plans. Its 1,000-word proposal applies to an applicant's own idea, not automatically to this advertised project. Confirm the actual form requirements; contact doctoraladmissions@bath.ac.uk for programme questions. Programme application guidance

Bath encourages applications from under-represented groups. Applicants may briefly explain circumstances that affected educational attainment in a paragraph at the end of the personal statement. If an applicant declares a disability, the Disability Service will contact them to discuss support.

URSA covers Home or Overseas tuition, maintenance and up to £1,000 per year for training. Training money is not annual disposable living income. The Engineering policy's up to 3.5 years is a general maximum, not a confirmed personal 42-month award. A four-year academic doctorate therefore does not imply four funded years. Overseas awards are capped without a verified percentage. The project notice's £21,805 is a 2026/27 reference, not a confirmed 2027/28 personal stipend. UKRI's corresponding rate also cannot establish this applicant's award. URSA terms, UKRI training-grant guidance

Fees, Living Budget and Upfront Costs

The published 2027/28 Overseas Engineering research fee is £31,350 annually; continuing fees may rise by up to 8% annually, and deferred entry can change the basis. This is underlying exposure if funding is absent or ends, not an extra bill where an award fully covers fees. Obtain written confirmation of maintenance, months, indexation, fee increases, continuation and any writing-up period before acceptance. 2027/28 Engineering fees

Bath's 2025/26, historical budget gives £405 weekly or £21,060 annually. Its displayed total includes a £37 off-campus utilities line, even alongside a halls scenario; if the actual halls contract already includes bills, a conditional no-double-count calculation is £368 weekly, not a second university quotation. Average rent is £230 weekly on 51-week terms; an off-campus reference is £750 monthly excluding bills. Check the remaining week, deposit, contract terms and actual 2027 prices. Doctoral accommodation is not generally guaranteed. Historical postgraduate budget, Doctoral funding and housing FAQ

Student visa and IHS costs are the applicant's responsibility. The October 2026 references are £558 for a Student visa and £776 per year for the student IHS, with the total determined by visa duration and part-year charging. These are not guaranteed 2027 prices. CAS, funds and permission remain individual requirements; do not infer a reliable monthly surplus from a historical stipend or treat health charges and relocation as covered training costs. Student visa, IHS charges

Applicant Fit and Preparation

The project requires a first or upper-second-class degree or equivalent, a science, mathematics or engineering background, and competent coding and machine learning; a Master's is advantageous. The general programme also lists a relevant Merit Master's route, but it does not automatically waive project-specific subject or skills requirements. Evidence of reproducible ML work, spatial data reasoning and critical benchmark interpretation is more persuasive than naming many model families.

A Chinese applicant has a positive route through worldwide equivalent-qualification admission and URSA's Overseas eligibility, subject to assessment. The China research-degree indexed guidance and the currently displayed direct page differ in coverage; a definitive current PGR percentage remains unresolved. Do not apply undergraduate/taught-postgraduate percentages as a PhD cut-off, or imply that a Master's automatically resolves equivalence. Ask admissions to assess the actual institution, degree, grades and research preparation. International qualification guidance

English is IELTS 6.5 overall, each component 6.0. TOEFL taken before 21 January 2026 requires 90 overall, each component 21; on or after that date, 4.5 overall, each component 4.0. Results must fall within the relevant 24-month period before entry. A qualifying English-taught degree may support an individually assessed exemption, generally within two years, with specified exceptions; a Chinese medium-of-instruction letter is not automatic approval. Crucially, non-UK applicants not currently studying in the UK must satisfy English before the funding deadline, with no exceptions. An exemption needs confirmation in time, not an assumption that an eventual offer condition suffices. Doctoral English requirements

The programme is ATAS-covered. China is not on the nationality-exemption list, but the actual requirement and subject classification must follow the offer/CAS and immigration advice. This is an individual clearance requirement, not a categorical nationality ban or a guarantee of approval. Prepare transcripts, research examples, references, English evidence and a focused fit statement; ask what additional documents this advertised project needs rather than automatically submitting an own-idea proposal. ATAS rules

Next Actions and Questions to Verify

Before applying, establish the first research question, actual city/feed rights and whether the initial contribution concerns tracking, spatial representation or forecasting. Ask how uncertainty, occlusion and cross-scene generalisation will be evaluated, and how train/validation/final tests will be separated. The RoundaboutHD split, dataset access and licence need inspection before reuse; no misconduct conclusion follows from uninspected metadata.

Before interview or acceptance, confirm the named daily supervisors, available compute/storage, annotation support, publication/IP conditions and personal training plan. Obtain the actual 2027 stipend, supported months, full fee coverage and increases, and support for completion beyond the scheme maximum. Resolve Chinese degree equivalence, English approval by the hard deadline, ATAS/CAS and alternative-start conditions. Finally, budget 2027 accommodation, upfront immigration costs and relocation separately. These unresolved details justify careful verification. The saved materials describe a formal application route, but current acceptance status remains unverified.

Evidence and Sources

These sources support the research, training and application assessment. Individual reading scopes are stated above; three complete-paper audits, code replication and dataset-licence inspection are not claimed.

Project, Team and Original Research

• Nic Zhang current Bath institutional profile — Current roles and publication metadata.
• Bath Centre for Spatial Intelligence research themes — Research themes, not guaranteed resources.
• Centre for Spatial Intelligence current staff — Formal affiliation, not appointed co-supervision.
• City-scale multi-camera vehicle tracking with improved self-supervised camera link model — Selected accepted-author Methods/Results/Conclusion; not final publisher full text.
• Ablation study for multicamera vehicle tracking using CityFlow dataset — Complete original institutional abstract and metadata only.
• RoundaboutHD urban multicamera benchmark author preprint v2 — Selected v2 Methods/Results/Conclusion; author preprint, no data/code audit.

School, Programme and Training

• Bath Mechanical Engineering full-time PhD programme — Programme, training, admission and application rules.
• Bath Mechanical Engineering research environment — Departmental context, not placement or salary evidence.
• Bath doctoral professional development — Institutional development support.
• QS Bath overall 2027 table — Overall institutional ranking.
• Bath postgraduate research school context — Institution-wide doctoral/research context.

Funding, Application and Eligibility

• Current University Research Studentship Award policy and 2027 Engineering deadline — Current Engineering competition; general maximum, not personal award term.
• Bath doctoral English rules — Tests, two TOEFL scales and assessed exemptions.
• Bath Chinese research-degree qualification route — China equivalence requires individual confirmation.
• Bath Engineering research tuition 2027/28 — 2027/28 fees and continuing increases.
• Bath postgraduate budget 2025/26 — Historical budget; utilities overlap must be checked.
• Bath doctoral admissions FAQ 2026/27 — Historical maintenance reference and accommodation boundary.
• UKRI training grant current stipend — 2026/27 rates, not a personal 2027 award.
• UK Student visa official route — CAS, application and dated fees.
• UK immigration health surcharge — Duration-dependent charges.
• ATAS current nationalities and subject rules — Nationality exemptions and individual subject clearance.