|
| Model name: | Traffic | Model title:
| EMIL-S Traffic Model | Replicated model?
| No | | Keywords: | Norm emergence, normative agent, car, pedestrian, collision avoidance, demonstration model, animation | Model authors:
| Ulf Lotzmann | | Programming language? | Java, EMIL-S |
The traffic scenario is one of the stylised fact scenarios, as the implemented simulation models reflect what the ordinary car driver or pedestrian will have in mind when describing and analysing the (normative) behaviour of traffic participants. It consists of several simple scenarios in which software agent learn from experience and observation which kinds of behaviour in street traffic are dangerous and / or unwelcome (as dangerous behaviour will not always need to be explicitly forbidden, and as sometimes behaviour is explicitly forbidden – or at least blamable – even if no danger arises from this behaviour). DOCUMENTATION 1. Purpose The field of microscopic traffic simulation is dominated by agent-based approaches which are focusing on precise mathematical modelling of physical parameters. In many cases, the interaction between traffic participants is restricted to measuring and keeping distance with the aim to reproduce realistic traffic flows while avoiding collisions (e.g. in classical car-following models). Psychological aspects of traffic participants as well as social relations between them are much less treated in current approaches, although these are crucial factors for the dynamics of any real traffic system in which usually various kinds of traffic participants (drivers, cyclists, pedestrians …) appear. In particular, social capabilities are the key for integrating differing perspectives (of distinct kinds of traffic participants) on a joint event.
Normative behaviour and learning introduce social capabilities for agents in traffic simulations. The following model is a step toward this more comprehensive view on traffic systems.
The inclusion of this model as an EMIL scenario has an additional purpose: Due to the fact that basically all potential users should be familiar with traffic matters in everyday life, it seems beneficial to call on such kind of scenario for the technical introduction to handling and integrating EMIL-S. 2. State variables and scales In this model two groups of agents – pedestrians and car drivers – are situated in a shared topographical environment. The topography is composed of a straight one-way road and two meadows to the left and right of the road. A small segment of the road has a special mark (much like a crosswalk). The pedestrian agents are allowed to access both the meadow and road regions of the topography and are able to perform the following activities: - move around randomly within an meadow;
- approach a special target agent situated within the meadow on the opposite side of the road, thus implying the necessity to cross the road;
- select a road section for the crossing procedure;
- modify the velocity (fast, slow, stop);
- select between perception modes (narrow, wide).
The car driver agents only move within the road regions of the topography and are able perform the following activities: - drive along the road;
- modify the velocity (fast, slow, stop);
- select between perception modes (narrow, wide).
3. Process overview and scheduling
Situated within one of the meadows, a number (which is constant during a simulation run) of pedestrian agents move around randomly. As soon as the target agent appears on the other meadow, the pedestrians begin with a “reach target procedure” with the aim to arrive at the target as fast as possible. For this activity, the agents can choose between the direct way to the target or a (possible) detour via a selected specific road section. The road is populated by car agents who are aimed at reaching the end of the road faster than the average duration.
For both types of agents, the deviation from the average duration leads to a valuation of the recent agent activity: a penalty when more time was required and accordingly a gratification when the target was reached earlier.
Due to the interaction between agents, occasional collisions are likely to happen. Such an event, when occurring between a car and pedestrian, is classified as an undesirable incident. Observations of a collision provoke other agents to issue sanctions against the blamable agents.
4. Design concepts
4.1 Emergence Emergence of norms within the groups of pedestrian as well as car driver agents to minimize the number of collision incidents. As a typical situation after a sufficient number of interactions (which is usually the case after a few hours of artificial simulation time, equivalent to several minutes of real time, depending on hardware) the pedestrians have learnt that they have to use a special (striped) area for street crossing (in spite of a potentially longer route), car drivers have learnt that they are expected (obliged) to slow down or stop in front of the striped area (which has emerged into an institution after the first successful norm learning happened there) when there are pedestrians visible in the neighbourhood. 4.2 Adaptation Agents adapt their behaviour by changing their action preferences, based on received feedback from the environment (perception of collision, reaching the target at an earlier or later time than the average) or received messages from other agents (punishment). Beside learning from own experience agents can in principle also learn from other agents’ actions.
4.3 Fitness The fitness of an agent relates to its norm compliant behaviour. 4.4 Prediction No prediction in this model. 4.5 Sensing Complex sensing (environmental perception) mechanisms are involved within the physical layer of the agent models: - perception of neighbouring agents of both types (including physical attributes like velocity);
- perception of topographical regions (several road section, meadow);
- perception of collision incidents and discrimination of the involved opponents.
4.6 Interaction Obviously, the model includes several possible (physical) interactions, from which the collision between a pedestrian and a car is the most important.
Observations of collisions provoke other agents to issue sanctions against the blamable agents. The extent of the sanction is determined by various factors reflecting the environmental situation (e.g. the road section where the collision occurred) and the normative beliefs of the valuating agent (e.g. a collision on a special section, e.g. crosswalk, might result in a harder sanction than on the rest of the road). Sanctions lead to a temporary stop of motion for the involved agents. Hence, to avoid sanctions is a competing goal to the general aims (reaching the target point or end of the road, respectively, in due time). 4.7 Stochasticity The actions are selected using a stochastic process. The initialization of action probabilities can be done randomly.
4.8 Collectives The model includes two agent groups: pedestrian and car drivers. 4.9 Observation The simulation process is visualized by animation sequences. Furthermore, the recording and statistical analysis of simulation variables is possible in principle. 5. Initialization Dependent on the experiment style, the agents are initialized with either uniform action selection probabilities, or with some biased actions.
6. Input The input parameters of a simulation run are: - number of agents per group, default: 20;
- configuration of biased actions;
- strength of sanctions.
7. Submodels No submodels. |
