WP-I3: Further development of the multi-agent system module: intelligent routing and human behaviour, evacuation simulations, and interface to MATSim for agent-based vehicle emissions

Goals of the project:

A multi-agent system (MAS) for pedestrians in urban environments was developed within the first funding phase of MOSAIK. Within the work package WP-I3.1, it will be combined with a vehicle-agent system (VAS) to simulate vehicles and their emissions in a more explicit way. In a first step, the VAS will be coupled to the external agent system MATSim (WP-I2) and will receive information about movement patterns and emission rates (point sources). In the second step the VAS will be coupled directly to the street network within PALM-4U and will allow an autonomous (i.e. independent from MATSim) routing considering the street network to be able to conduct idealised simulations of traffic emissions (e.g. for different variations of the vehicle fleet within an urban canyon using standard emission rates). In the medium term, it is planned to couple the developed VAS to a vehicle model so that the traffic-induced turbulence can be taken over by the model. The development of the vehicle model is part of a current DFG research project (MA 6383/3-1), therefore synergy effects can be expected during the upcoming 2-3 years.

The main task of WP-I3.2 is the coupling of the MAS to the street network, based on OpenStreetMaps?, which has already been implemented in PALM-4U for the demonstration cities (Berlin, Hamburg, Stuttgart) and is currently in use for default traffic emissions. The concept of „social forces“, which is used for the MAS, will be further developed for the application on streets. Streets – depending on their classification – serve as a repelling force on the agents. By doing so they are forced in the nominal case to move on the sidewalks. Information from OpenStreetMaps? regarding pedestrian crossings and traffic lights will be used to define permeable areas where agents can cross the streets.

In WP-I3.3 the MAS will be extended in regards to including demographic data. This affects in particular the walking speed. Demographic input data and movement patterns (shaped as start target matrices) for the entire Population of Berlin and Stuttgart will be provided by MATSim (WP-I2) and serve as input data for MAS. Test simulations for a heatwave scenario will be conducted and biometerorological parameters for individual agents will be calculated and analysed to filter out the demographic effect on these parameters. The final version of the MAS will be provided to the partners of module C for operational applications.

The individual exposure of agents to air-borne pollutants will be added to the MAS in collaboration with WP-I1. Concludingly the MAS will be applied to different test cases in so-called emergency scenarios. Here, the MAS will be used in reversed mode, i.e. the agents are escaping on the fastest way from a defined hazard (fire, explosion, toxic pollutant source). PALM-4U will provide the wind field and therefore the dispersion of toxic material which can affect the agents on their escape route from the hazard source. At the same time, the agents will be influenced by social forces (e.g. slower walking speed due to high agent density) and might be exposed to the pollutants. These simulations have a very practical character as existing escape route concepts in cities can be evaluated and improved, for example for sports or large concert events. The applicability will be evaluated in addition with external partners from module B (Prof. Ament, Prof. Leitl, University of Hamburg).

Tasks of the project:

WP-I3.1: Development of VAS system and coupling to MATSim

WP-I3.2: Coupling of the MAS to the street network

WP-I3.3: Adding demographic data and movement patterns

WP-I3.4: Coupling to chemistry and emergency simulations

Projekt structure:

This work package will be conducted in close cooperation with TU Berlin (Prof. Kai Nagel).

Deliverables:

DL1 (month 12): VAS component is implemented and coupled to MATSim output data

DL2 (month 21): MAS is coupled to the street network

DL3 (month 27): MAS is ready for application by Module C partners for demo cities

DL4 (month 29): Pollutant exposure of agents is implemented

DL5 (month 36): Emergency simulations were performed

Progress so far:

The multi-agent system (MAS) for pedestrians in urban environments, which was developed within the first funding phase of MOSAIK, is used to generate realistic movement paths for pedestrians. An improvement for the better representation of the pedestrians has been developed: the new demographic population compositions allow a more realistic approach to movement dynamics, based on randomly age, gender, height, weight, activities, etc... (e.g. elderly people and children will move slower, height and weight of the children are smaller). Another improvement takes into account the movements of the pedestrians on very well defined footpaths, first assuming that only 2 grid cells around all buildings should be used for side-walks (using Minkovski sum between the polygonal building and the shape of side-walks) and after that based on “street_types” definition in static_driver some of them can be shared with vehicles. Finally, crossroads are added for connecting different available paths. Therefore, the new navigation mesh contains information only about the available area for agents and they try to not "run wild".
There are still some limitations of the MAS:

• Agents can "run wild" if the "static_driver" have no information about "street_type", or "crossroads",... (the new navigation mesh cannot be generated, and only the buildings are seen as obstacles)
• Extra load moving on a slope is ignored
• No feedback from the environment (sun/shadow, pollution)
• Agents do not measure chemical compounds or aerosols. This extension is in preparation.

The MAS source code is used as the basis for a new Vehicle Agent System (VAS) module in order to couple MATSim output data (i.e. emissions and movement paths) to PALM-4U. For this, a new coupler MATSim-PALM-4U was created to read the movement paths and traffic emissions of the vehicles from MATSim output and incorporated them into the PALM-4U. Since the time step of the vehicles from MATSim output files (1 second) is larger than the model time step of PALM-4U, the new module of VAS will contain an interpolation/accumulation procedure of the vehicles' emissions for each grid cell of PALM-4U. The interpolation algorithm computes each cell visited by the moving vehicles within the PALM-4U grid and then calculates the fraction of time the vehicles spent in each grid cell. Based on this time fraction, the vehicles' emissions are updated on every grid cell visited by the vehicles.

The algorithm also treats some special cases, when the vehicles are moving exactly on the grid lines (horizontal or vertical) and on the edges of the domain. There are some assumptions/modifications to the algorithm, such that:

• the vehicles moving inside the domain, exactly on the grid lines, are releasing emissions always on the left side.
• the vehicles moving on the edges of the domains are releasing emissions always inside of the domain, regardless of the direction of the movements.

Also, the pedestrians' movements from MATSim output have been taken into account in the new interface: each initial and target position of the pedestrians, and starting and ending time of the movements of the agents were incorporated in PALM-4U. Then the main model runs the MAS with the new agents for the given simulation time.

References:

Hershberger, J., Snoeyink, J., 1994. An O(nlogn) implementation of the Douglas-Peucker algorithm for line simplification, SCG '94 Proceedings of the tenth annual symposium on Computational geometry, 383-384.

Helbing, D., Molnar, P., 1995, Social force model for pedestrian dynamics. Physical Review E, 51(5), 4282.

Karamouzas, I., Skinner, B., Guy, S.J., 2014, Universal Power Law Governing Pedestrian Interactions, Physical Review Letters, 113, 238701.

Contacts:

maronga[at]meteo.uni-hannover.de

raasch[at]meteo.uni-hannover.de

matei[at]meteo.uni-hannover.de