Dr. Nicolas Verstaevel

Homepage

Research

Ongoing projects

http://www.irit.fr/MIMICO

The ANR-JCJC MIMICO (2024-2027) project is funded by the french national reserch agency (ANR) under the AAPG2024 funding scheme. MIMICO is a young researcher project led by Dr. Nicolas Verstaevel, with the aims to provide next genration machine learning technics applied to agent-based modelling.

Agent-based models (ABMs) are interesting tools for modelling and studying complex phenomena in which numerous heterogeneous entities with non-linear interactions are geographically distributed and modelled at different scales. Scientists interact with ABMs by modifying the values of model parameters, either during a calibration process to produce realistic data, or to explore possible outcomes in ‘what-if’ scenarios. In-depth exploration of the parameter space of an ABM is difficult due to the relatively large number of parameters, the potentially high computational costs per model run and the non-linear relationship between parameters and results. The MIMICO project proposes to design and evaluate a new approach based on the construction of a substitution model to emulate the relationship that exists between a scenario and the results of the agent model. The novelty of the project is to design new active learning strategies, which implies that the substitution model will itself observe its learning process to request new examples. The key hypothesis is that if the learning algorithm is allowed to choose the data from which it learns, and therefore to be curious, it will perform better with less training. The surrogate model can then discover new scenarios to explore, providing interesting information for the expert while improving its own ability to mimic the ABM. The framework is being evaluated in two application areas: the first is the evaluation of the impact of new urban policies on the mobility model, and the second is the calibration of dense crowd simulation.

The French-German project MADRAS

Project MADRAS aims to develop innovative agent-based models to predict and understand dense crowd dynamics (from 2 to 8 ped/m²) and to apply these models in a large-scale case study.

Indeed, trustworthy models for the dynamics of dense crowds are crucial for the prediction of pedestrian flows and the management of large crowds, but also from a fundamental perspective, to understand the roots that they share with active matter but also the pedestrian specifics. However, current models suffer from some severe deficiencies, especially at high density. Two complementary modelling approaches will be pursued: 

(i) neural networks (NN)

(ii) a physics-based model coupling a decisional layer.

In adddition, emulating a real scenario calls for adequate data assimilation methods and efficient multi-agent simulations based on the two models. These models will be combined in a single online platform, allowing one to visualise the predicted flows and compare them with the ground truth. 

Finally, the impact of different model ingredients and features (e.g., shape heterogeneities) on the large-scale flow predictions will be investigated by means of numerical simulations of the two models, using the Festival of Lights situation as a reference scenario.

Four teams with different backgrounds (Computer Science, Statistical Physics, Applied Mathematics) combine their strengths and research tools (simulation platform, continuous models, datasets, experimental analysis tools) in order to achieve this ambitious interdisciplinary project:

ANR SWITCH

Transport infrastructures play a large part in defining the city of the future, which should be smart, sustainable and resilient. Their management will need to deal with the emergence of novel technologies (i.e. autonomous cars, Internet of Things), the apparition of novel modalities (hoverboard, etc), and changes of practices (increase of multi-modality, electric bicycles, shared cars). These aspects could favour and accelerate the transition to the city of the future with positive social, environmental and economic impacts, in order to address foreseen trends (climate change and new requirements in terms of pollution, security, and global costs).

The SwITCh project seek to integrate a large variety of urban transport modalities (private car, walk, tramway, bicycle, etc.) and associated infrastructures (pavement, tram track, bicycle path, etc.). It aims to support decision-making for urban planning by simulating the gradual introduction of disruptive innovations on technology, usage and behaviour of infrastructures. Achieving such an objective requires providing a model that is able to assess the impact of these innovations on several key indicators related to mobility, user satisfaction and security, economic costs, and air pollution. It requires a holistic vision of urban transport to avoid indirect consequences of choices or rebound effect (e.g. a solution improving air pollution but reducing security). The model must include current and future infrastructures and modalities, and consider the transition process between current and future situations. We do not provide an a priori list of potential innovations; some of them are already considered (e.g. autonomous car, smart infrastructures, car sharing, etc.), but others will appear from the WP1. Being able to consider a large variety of disruptive innovations and indicators requires a high flexibility of the model; this issue will be at the centre of the modelling process.

The consortium is composed of 5 research laboratories from different domains (engineering, geography-urban planning, computer science, etc.) :

RECOVER/IRSTEA: The RECOVER team gathers researchers with multidisciplinary skills in environmental sciences, engineering, ecology and decision support systems, to study the interactions between natural and manmade systems for risk management. The team aims at developing knowledge, methods, tools and information systems to analyse risks and vulnerabilities in order to create an integrated management system for hydraulic works, territories and their ecosystems. The involved researchers have a strong expertise in decision-making for infrastructure management in uncertain contexts.

I2M/GCE: The Civil and Environmental Engineering Department (GCE) of the Institute of Mechanics and Engineering Bordeaux (I2M) deals with various issues related to the field of Civil Engineering. It develops research on engineering respecting two key elements: (a) a very strong interdisciplinarity, combining various approaches (physical modelling and decision support) and dealing with problems on a wide variety of scales; (b) the constant concern to build a continuous and balanced chain, from the knowledge and the modelling of the phenomena/mechanisms involved upstream to pre-operational conclusions, which can be seized by users (industrialists, engineering offices, real property managers, etc.). The GCE team participation in numerous national and international projects demonstrates its deep involvement in the civil engineering field.

LIG/HAwAI: The LIG-HAwAI team (Human Aware Artificial Intelligence) is focused on creating socially relevant multi-agent systems that solve real-world problems and enrich society. HAwAI was previously known as MAGMA, one of the oldest multi-agent systems teams in France. HAwAI is part of the 23 teams that make up the Grenoble Informatics Laboratory (LIG), one of the largest computer science research laboratories in France. The HAwAI team extends the notion of an agent-based society away from a collection of purely computational entities, to one that is composed of different types of agents, both human and artificial, interacting in mutually beneficial ways. In particular, the team is focused on modelling social interactions and simulating human behaviours. Through agent based simulation the team has previously looked at analysing human mobility and crowd movements in earthquakes (ANR-LIBRIS, VUSIM AMD, ARC7 MIMICS, IXXI MUDAMO), in floods (CNRS MAGIL, IDEX CDP RISK), and in bushfires (AGIR-SWIFT), with a focus on human factors such as emotions, cognitive biases or social attachment.

IRIT/SMAC: The Cooperative Multi-Agent Systems (SMAC) team of the Institute of Research in Computer Science of Toulouse (IRIT) is one of the most important teams in Europe working on Multi-Agent Systems (MAS), with applications in industry as well as research projects ranging from optimization to user profiling, including agent-based social simulation (ABSS). In particular, the researchers involved in the SwITCh project have a strong experience in ABSS in a multidisciplinary context. They are involved in several important research projects and teaching on the subject. They work on elaborated architectures for the design of realistic behaviours (ANR-ACTEUR), on the generation of synthetic populations (ANR-Genstar) as well as on the modelling and analysis of networks dynamics using Time-Varying Graphs and MAS.

ThéMA/MCT: The ThéMA/MCT team endeavours to understand the spatial dynamics and structural links between accessibility, mobility and distribution of activities. The SwITCh project could benefit from their previous results and methodology of two projects. The first one concerns the Miro platform, which combines activity theory with an agent based model. The second one concerns the Mobisim Project. This Land Use Transport Integrated Platform simulates and analyses geographical daily and residential mobility dynamics. It supports decision-making for sustainable planning of French and European cities.

PhD Supervision

In collaboration with TwinsWheel

 The aim of the thesis is to improve the safety of autonomous mobile droids that intervene in the presence of road users (pedestrians, cyclists, children, etc.). These autonomous vehicles are equipped with artificial intelligence algorithms to make decisions on the trajectories to follow, the obstacles to avoid, and the behaviour to adopt. It is important to put in place tools to ‘monitor’ the behaviour of the droids in order to detect any anomalies by integrating tools and methods from the partner laboratories.
The aim of the thesis is to propose approaches based on Multi-Agent Systems to build surrogate models capable of providing explainable detection models.
For TwinswHeel, demonstrating its ability to integrate this type of security as soon as possible will give it a major competitive advantage.

Funded by the ANR-DFG MADRAS

Trustworthy models for the dynamics of dense crowds are crucial for the prediction of pedestrian flows and the management of large crowds, but also from a fundamental perspective, to understand the roots that they share with active matter but also the specifics properties of each pedestrian. However, current models suffer from some severe deficiencies, especially at high density.
Trustworthy models for the dynamics of dense crowds are crucial for the prediction of pedestrian flows and the management of large crowds, but also from a fundamental perspective, to understand the roots that they share with active matter but also the specifics properties of each pedestrian. However, current models suffer from some severe deficiencies, especially at high density.
In this context, the MADRAS project (funded by French ANR and German DFG agencies) aims to develop innovative agent-based models to predict and understand dense crowd dynamics (from 2 to 8 ped/m2) and to apply these models in a large-scale case study. Two complementary modelling approaches will be pursued: (i) neural networks (NN) that will be trained on available data to predict pedestrian motion as a function of their local environment and trajectory, (ii) a physics-based model coupling a decisional layer, where a desired velocity is selected according to an empirically validated collision-anticipation strategy, and a mechanical layer, which takes care of collisions and contacts. To push this approach to higher densities, integrating more realistic pedestrian shapes and better splitting the decision-making process from mechanical forces is necessary. These approaches will be confronted with novel validation methods, using data from controlled experiments and extended to a large-scale real case study (the Festival of Lights in Lyon).
The objective of this thesis is to exploit the two models developed in the project at a larger scale to simulate the flows on crowded streets at a real mass gathering, the Festival of Lights in Lyon, and to validate it with the temporal gathered data. To this end, empirical data will be collected by filming the streets from above and by immersing in the crowd participants wearing pressure-sensing jackets, to measure contacts.
Emulating this real scenario will call for adequate data assimilation methods and efficient multi-agent simulations based on the two models. This model will be developed in a single agent-based platform, the generic open-source agent-based modelling and simulation GAMA platform.

In collaboration with Citec

Pattern detection on massive data is one of the standard problems addressed by Artificial Intelligence in general. The reconstruction of flows that depend on the detected patterns and profiles and the ability to take into account both their distributed character as well as their spatial and temporal evolutions is a particular stake of the thesis. Innovative approaches in artificial intelligence address these issues. This is the case of Multi-Agent Systems which allow a design based on the analysis of the business domain by defining autonomous entities with a local vision of the system, and which interact to globally define the appropriate function.

Ministerial funding

This thesis aims to propose a cooperative agent model, based on the self-adaptive multi-agent system theory (AMAS), allowing an efficient and fast exploration of the parameter space, autonomously and automatically. This exploration should allow a continuous readjustment of the simulation until convergence, improving the control of the macro-level over the micro-level. On an application standpoint, the purpose of this project is to produce a realistic traffic that satisfies the best a set of objectives and constraints at both micro and macro levels. This traffic should also allow interaction with humans and adapt to events that could occur in the virtual environment.

 

Funded by ANR SWITCH project

The ANR SWITCH project objective is to study the impact on transport modalities of new infrastructures for Smart Cities. Those new modalities of transportation (electric vehicle, autonomous vehicles, mobility as a service, bikes, ….) can facilitate intra-urban mobility or improve quality of life, but they can also create new constraints for which cities must be prepared. Urban planners need a new generation of tools to assess the impact of urban policies in terms of mobility and infrastructure to explore “what-if” prospective scenarios. The new interactive simulation tools require to simulate mobility and it’s impacts at the scale of the city over different timescale (from 1 minute to 10 years).

Agent based simulation is an interesting tool to model complex phenomenon such as traffic (Bazzan,2014). Large scale road traffic is a complex system: it is composed of numerous heterogeneous entities with nonlinear interactions that are geographically distributed, and can be modelled at different levels (Haman, 2017). A microscopic simulation focuses on modelling each vehicle and its dynamics are modelled individually, whereas a macroscopic simulation focuses on aggregated information such as traffic density and traffic flows (Lopez,2018). To study the impact of urban planning decisions, it is necessary to design tools that can combine and switch between those different scales at runtime (Haman,2017). Furthermore, it is also interesting to combine those different models with other models to study for example the impact on environmental factors (Schulze, 2017), or the individual response of citizen to catastrophic events (Chapuis,2018). Therefore, designing multi-level agent-based simulations addresses different challenges in term of software design (Mathieu, 2018), the coupling of models with different input/output and dynamics (Taillandier, 2019), and complexity of the model when millions of heterogeneous entities must be computed (Mastio,2018).

Different agent-based simulation tools exist in the literature, each tool having pros and cons in terms of computational modelling capacities and scalability (Abar,2017).

The thesis will design and evaluate a new metamodel for interactive Smart City simulations where different models can be combined and switched at runtime. The metamodel will be applied to evaluate the impact of different mobility evolution scenarios in the French cities of Dijon and Bordeaux. The thesis will contribute to develop and extend the open-source agent-based simulation platform GAMA (Taillandier,2019).