Amble - a pedestrian traffic behavior simulator

This research offers several new contributions to the area of pedestrian movement simulation:

  • Group awareness simulated with extensions to computer graphics "flocking" work.
  • Representation of information in the environment and within pedestrians.
  • Ability for pedestrians to learn from environment and schedule activity.

    First, a quick note about simulating movement. Commercialized pedestrian movement software contains at least as much eye candy as simulation. 3D models and environments add little to the simulation but of course are a necessary part of public presentations. Amble is a research tool and contains none of that, but don't think less of it! Take note of other simulators, in your mind's eye strip them down to the simple dots displayed in Amble, and see how many support group interaction and learning from the environment, both of which are very real parts of pedestrian movement that affect use of public ares. It is unlikely (in 2012) you will find any because of the difficulty in addressing the topics. Now imagine what a team of computer scientists could do with 3D graphics to bring Amble to life for public presentations and usage studies.

    The links to the left offer examples using Java applets of increasingly complex pedestrian movement models. When basic walking looks acceptable, we then add higher level decision making like scheduling and planning to each pedestrian. Finally, we add simple learning. Each inanimate agent, e.g., a store, offers information to nearby animate agents (pedestrians). The pedestrians use the information to further focus their schedules. For instance, a sign might indicate to a pedestrian that Platform 2 is this way. With the new knowledge the pedestrian proceeds in that direction. At this point, with reasonable data regarding pedestrians and destinations, the simulator can offer reliable predictions of use to designers and planners.

    A little more detail

    For flexibility in trying new ideas and further researching old ones, agent description and behavior are not hard coded into the collection of Java applets on this page. A simple formal language is used that describes agent properties, environmental properties, and things that can be learned. This makes it possible to easily pause the simulation, save its state, and reload it later; not so important in these simple demonstrations, but important in large, complex ones. Also, the fact that general descriptions and "atoms" of knowledge are indentical in structure makes it possible for pedestrians to both learn from and deposit information into their environment. When told, for instance, to take a train to Haslemere, a pedestrian can use information in the environment to first find the station and then the platform. An agent can similarly affect his environment by sharing learned information with others. In addition, an agent's properties can also affect the actions of other agents. Poll takers might repulse passersby while a "free coffee" sign will attract some.

    Still curious?

    If you still want to know more after reading the above overview, then you just might be interested in the dissertation itself. The pdf version can be downloaded here:

    Modeling Behavior in Vehicular and Pedestrian Traffic Flow

    Feel free to drop me a line!

    Mike Markowski
    mm@udel.edu

    website
statistics