MSc Thesis Proposal
Multi-agent multi-class traffic flow control
Bart De Schutter
Keywords: traffic control, optimization, multi-agent
On the short-term dynamic traffic control is the most viable
option to address traffic congestion problems.
In this project we will
study a distributed control set-up where local controllers
(called agents) optimize the traffic control strategy for their own
area (which could be a single intersection, a number of intersections,
a part of the urban network, an interface between the urban and
motorway network, or an entire ringroad). The interactions between
areas governed by different autonomous agents have to be considered in
order to avoid that the system consists of a collection of locally
optimized processes, instead of a network-wide optimal
situation. Since the control actions of one agent may have negative
effects on the traffic situations controlled by other agents, some
form of coordination of their control actions is required. The study
of this coordination problem is one of the main subjects of this
subproject. Coordination can be effectuated by communication and
negotiation between the different autonomous agents, or by an
hierarchical control approach, where coordination is done by a
higher-level supervisory controller (see figure).
Figure 1: A schematic representation of a hierarchical multi-agent
traffic control framework.
The agent-based approach has
several advantages compared to a centralized control set-up:
distributed control is more flexible and robust to individual failures
( graceful degradation ); it has good scaling properties, can be
extended more easily, and is also better suited for on-line real-time
traffic control applications.
The optimization of the interaction
between urban and motorway networks (e.g., avoiding jams on highways
due to clogging of urban arterials and off-ramps) is a focal point of
this subproject. In addition, special attention is deferred to
socially and economically important user-classes.
If you are interested in selecting this project as your MSc project,
please come along or send us an email for more information.
This page is maintained by
Bart De Schutter.