What we can do
Six competence groups shared by all five research pillars. Described once, here, so that there is a single consistent answer to the question a prospective partner actually asks.
Methods, not applications
A pillar is a research field. A method is something we can apply to your problem tomorrow. Keeping the two apart is why this page exists — and why you will not find the same capability described in five places.
Mathematical modelling and econometrics
Econometric methods and hypothesis testing. Discrete choice, hybrid choice and integrated choice–latent variable models. Latent-class and mixture models. Logistic regression. Dynamic and Hidden Markov formulations. Structural equation modelling. State-space modelling and model predictive control.
- Pillar 1
- Pillar 3
- Pillar 5
Soft computing and artificial intelligence
Fuzzy inference systems, artificial neural networks, ANFIS, genetic algorithms, multi-agent systems. Deep neural networks. Reinforcement learning as a benchmark rather than a default. SHAP and explainable AI. AI-supported systematic literature review.
- Pillar 1
- Pillar 2
- Pillar 3
Statistics and data processing
Bayesian identification and recursive estimation on discrete data. Data mining on loop-detector and floating-car data. Dependency analysis in small samples. Design and evaluation of surveys, including sociological surveys. ETL pipelines for heterogeneous municipal feeds; GTFS, APC and OSRM processing.
- Pillar 2
- Pillar 3
- Pillar 4
- Pillar 5
Operations research
Linear programming and its applications. Game theory, including non-cooperative equilibria and asymmetric Nash bargaining. Network optimisation and graph theory.
- Pillar 1
- Pillar 5
Mathematical modelling of traffic and traffic simulation
Microscopic simulation in Eclipse SUMO, to whose development we contribute, and in PTV Vissim and Aimsun. Multi-agent simulation in AnyLogic. Large-scale agent-based demand in MATSim. Multi-resolution meso and micro coupling. Environmental coupling to the PALM large-eddy simulation.
- Pillar 1
- Pillar 2
- Pillar 4
Impact assessment, validation and virtual testing
KPI definition and experimental design. Before-and-after evaluation. Impact assessment of policies and technologies by microscopic simulation. Statistical processing and defensible attribution of effects. Virtual test environments as a complement to physical testing of automated vehicles.
- Pillar 1
- Pillar 2
- Pillar 4
Impact assessment is a competence, not a by-product
It is the work package we led in H2020 MAVEN, and the reason the capacity, delay, queue and CO₂ figures from that project survived review and reached a leading journal. It is also, in our experience, the work package that consortia routinely under-resource and then cannot defend.
It sits in the methodological core rather than inside a single pillar because we apply it to demand policies and public transport measures as much as to automation.
Four generations, each kept accountable
Decision trees (2008–2010), then fuzzy inference (2012–2015), then multi-agent systems (2015–2019), then deep learning and explainability. Each generation was kept accountable rather than simply replaced — and in operational deployment the interpretable option sometimes still wins.
That is not nostalgia. It is why we can defend a tool choice to an operator who has to sign for the consequences.
Software and instruments
- Eclipse SUMO
- PTV Vissim
- Aimsun
- AnyLogic
- MATSim
- PALM
- TraCI
- OSRM
- GTFS and APC pipelines
- Python ETL
- Bayesian identification
- SHAP
- Driving simulator with eye tracking
- Instrumented vehicle
- OpenDRIVE / OpenCRG
Would you like this applied to your problem?
We work with cities, infrastructure managers, operators, suppliers and other universities — from a single analysis to a six-year consortium. See how we work with partners, or write to lambda@fd.cvut.cz.