Ontologies in multi-agent systems for building design. The case of risk management inside a stadium.

Ontologies in multi-agent systems for building designThe case of risk management inside a stadium

Matteo Caglioni (UMR ESPACE - Université de Nice Sophia Antipolis)Giovanni Rabino (DiAP - Politecnico di Milano)

COST Action TU0801 Workshop

3D Issues in environmental and urban systems

Computer Science School Technical University of Madrid (UPM)

12-13 April 2012, Madrid, Spain

12-13/04/2012, Madrid, Spain M. Caglioni, G. Rabino

2Contents

• The reasons of the raising interest of modelling pedestrian behaviour

• Ontology of pedestrian movements

• Case-study: an Olympic stadium

• Conclusions and further researches


3Modelling pedestrian movement. Why?

Pedestrian behaviour is the best example of the shift from macro to micro (meso) modelling attitude, because of:

more realistic models of collective behaviours

(e.g.: commuters flows, beyond the «gravitational» simplification)

possibility of modelling of previously intractable phenomena

(e.g.: inflows/outflows of a wagon of a «metro» train)


4Modelling pedestrian movement. Why?

In a evolutionary perspective (morphogenesis form function

relationship) a new focus on movement (urban) fabric link, where the agent (pedestrian) behaviour is crucial: movement defines (urban) stock configurations

(e.g.: self-organization of footpaths)

(urban) configuration defines movements

(e.g.: pedestrian movements in a «metro» station)


5Modelling approaches to pedestrian movements

Mainstream: Models based on kinetics/mechanical assumptions

Models bases on maximization of an utility function for agen

Here: Mainstream models are based on the real world. Instead our model is based on the perception (cognitive agent) of world

These models (analytic – e.g. queue models; or algorithm – e.g. cellular automata) apply bio-mechanical equations of pedestrian motion

These models (algorithm – e.g. system dynamics or MAS) solve a set of interdependent motion functions (e.g.. speed, direction, etc., for each agents) under a set of constraints


6Ontology of pedestrian movement (1/6)

Pedestrian movement is a complex adaptive process of (at least) 4 interacting psycho-mechanical factors (classes of the ontology):

• Orientation

• Path finding

• Routing

• Motion (walking, running, … )



Attributes of the classesand interactions:orientation

Agent action and mind

Interact with environment

Interact with other agents

3d model and semantic enrich

Updating the mental mapof locations

Looking around for landmarks; use

of cartography; …

Asking for information;reasoning by induction over

collective pedestrian behaviour ; …

Buildings LOD 1;D.T.M.;

enriched with info relevant for the moving

purpose



Attributes of the classesand interactions:path finding





Defining (updating) a

potential (i.e. on the mental map) desired* path to

destination

Reading the urban fabric as a “network” (dead-road, etc.); looking at the

weather conditions; …

(in case) following a guide;

asking for suggestions;

…

Buildings LOD 1 and D.T.M. read as a 3d

greed;enriched with info

relevant for the path choice

* According to mental (eg. being late), physical (eg. being tired) and environmental (eg. raining) conditions



Attributes of the classesand interactions:routing





Matching desired path with

real street network (and

their aspects and “traffic”

conditions)

Looking at many aspect of streets(shops, road –bed, road-signals, traffic congestion; …)

Supervised organization or auto-

organization of different “flows” of

pedestrian; …

Buildings LOD 3;enriched with info

relevant for advancing in the street (according to the path and the

purpose)



Attributes of the classesand interactions:locomotion (walking)





Doing displacement in a given condition (the specific site and time, the

“environmental” circumstances )

Paying attention to the situation

(slippery pavement, impending danger,

etc)

Physical (e.g. space occupancy) and cultural (e.g.

“personal” space ) interference; and/or vocal or non-vocal (e.g. glance) signs

Enrichment of building LOD 3 with details

essential for modelling displacements (e.g. location of pavement

slides)



Interactions among the classes

• Orientation

• Path finding

• Routing

• Motion (walking)

Classes (Psycho-mechanical factors) are simultaneous and embedded according to their specific spatial scale

BUT

there are many scale and time feed-backs (e.g. Learning processes) such as routing => orientation or routing => path finding or motion => routing


12Case-study: an Olympic stadium (1/7)

- 2 stadium configurations

- Temporary tribune module

- 2536 seats for each module


13

• Crowd and collective panic

Irrational behaviour of people

to guarantee, in the immediate,

their survival on detriment of the others

- Diffuse anxiety before disaster

- Occurring of a triggering event (meltdown)

- Lack of authoritative information

- Fast and progressive closing of the exits

Case-study: an Olympic stadium (2/7)


14

• Ontologies for semantic enrichment of project elements

Project elements: doors, stairs, seats, …

- Materials affect the people perception

- Stadium structure is a constrain for the movement

- Technological devices are useful to drive or regulate the flow of people

- Environmental conditions affect the objects and the agents inside the stadium



15

• Ontologies in multi-agent model for panic analysis

- Agent types: spectators, organizers, safety guards, firemen, rescuers, …

- Compression due to several agents can kill another agent (weakness)

- Signals (information) can be perceived by one or several agents

- Face to danger agents can act in different ways

- Emergency managers are cognitive agents, they drive people to accessible exits



16

• Characteristics of the system dynamic

- Agents move on a regular square grid (Von Neumann neighbourhood).

- In normal conditions agents move with calm speed (if not hindered).

- Agents are fatally compromised by triggering event and by compression.

- During panic people go towards exits.

- if they meet an emergency manager they follow his instructions.

• Calibration et validation

- No experimental data available for the stadium

- Data from the literature or likelihood value

- Data from others stadium cameras



17

• Model simulations (Monte Carlo approach)


Accident near the exit Accident in the north part

Initial state Panic diffusion Emergency managers


18

• Model remarkable results

- Exits guarantee outflows compatible with time defined by law (< 8 minutes)

- Simulation can show some unexpected situations

- Project exit configuration does not minimize collateral damages

- Replacing exits to extremes helps to minimise panic and injured people


Thanks for your attention

[email protected]

[email protected]

COST Action TU0801 Workshop

3D Issues in environmental and urban systems

Computer Science School Technical University of Madrid (UPM)

12-13 April 2012, Madrid, Spain

Ontologies in multi-agent systems for building design. The case of risk management inside a stadium.

Science

Transcript of Ontologies in multi-agent systems for building design. The case of risk management inside a stadium.