Agent Framework and Auction

In my market-based traffic controller, the intersection is composed of two types of agents:

intersection agents and traffic signal agents. At an intersection, there is a single inter-section agent and multiple traffic signal agents (see Figure 4.1). The interinter-section agent

Figure 4.1: The agent framework for the market-based traffic controller. The inter-section agents (also auctioneers) are responsible for making adjustments to traffic signal timings and executing auctions. Traffic signal agents, on the other hand, operate on behalf of a small set of legal vehicle movements that may occur at the intersection. The traffic signal agents compete against each other for control over traffic signal timing

adjustments.

is responsible for making adjustments to traffic signal timings and ensuring that those changes do not violate any basic traffic regulations (e.g., minimum green times). Traffic signal agents, on the other hand, operate on behalf of a small set of legal vehicle move-ments that may occur at the intersection. That is, each traffic signal agent is assigned a number of movements to manage. The traffic signal agents compete against each other for control over traffic signal timing adjustments. An intersection agent and its associ-ated traffic signal agents work together at the intersection level to adapt signal timings in real time. The adjustments are made to improve the efficiency of the intersection and maintain its safety.

The traffic signal agents are equivalent to traffic phases [22] in that they too rep-resent a set of vehicle movements. Thus, for every phase in the phase plan, there is a traffic signal agent that functions on its behalf to tweak the time allotted to that phase.

Together, all the phases form the signal timing for a traffic signal, while the traffic sig-nal agents function as an intelligent counterpart to the phase. These two constructs, phase plan and traffic signal agents, address the needs of all legal vehicle movements as traffic demands change. The design guidelines set by traffic engineers for phase plans (e.g., in the U.S., they use MUTCD [18]) therefore provide a blueprint for determining which movements will be assigned to which traffic signal agent. Figure 4.2 illustrates the relationship between the traffic signal agents and the traffic phases. As there are two phases, there are also two traffic signal agents.

There is a natural conflict that arises between traffic signal agents assigned to an intersection. Each traffic signal agent is designated to a single phase in the traffic signal timing. They compete for a slice of the limited amount of available green time in a cycle (see Figure 4.2). Assuming the cycle length remains the same, giving more green time

Figure 4.2: Traffic signal agents and their relationship to phase plans. For each phase, their is a corresponding traffic signal agent.

to one traffic signal agent means taking it away from another traffic signal agent. Thus, a multi-agent interaction protocol [13] is needed to determine an appropriate, adaptive allocation of green time to two competing entities.

As traffic flows through the intersection, auctions take place at fixed intervals which is called the auction period. The intersection agents serve as auctioneers and facilitate the auctions, that is, collect bids and determine the winner of the auction. The traffic signal agents participate in the auction and bid against each other to dictate how the traffic signal timing will be adjusted. The winner is the traffic signal agent with the highest bid. Note that the auction period does not have to match the cycle length.

An auction may occur in the middle of a cycle or after a series of cycles have passed.

Signal timings are only updated after the current traffic signal phase has completed.

As a safeguard against starvation, traffic signal agents are prevented from having less than 10 seconds of green time. Using the taxonomy described by Parsons et al. [62], the auction used in my work can best be categorised as single dimension, one-sided, sealed-bid, first-price and single-item. Thus, the manner in which the auction is utilised in my market-based traffic control system resembles the sequential single item (a sequence of single-item auctions) auction in market-based multi-robot coordination [12, 111].

4.2.1 Vehicle Detectors

The most common type of vehicle detectors are inductive-loop detectors. Inductive-loop detectors are coils of wire laid into the ground. A small current is passed through the

coils (or loop) to create an electromagnetic field. A vehicle (or any other large metallic object) that passes through the field will create a magnetic disturbance which signals a vehicle is present. Vehicle detectors may use any number of active and passive means to identify vehicle(s) within the detection zone, including video images, microwaves, lasers, radar, and infrared acoustics [112]. Vehicle detectors can be placed upstream from the intersection or downstream (near the intersection stop line). The traffic signal agents in my approach utilise vehicle detectors to assess road conditions and generate bids. Vehicle detectors provide estimations of traffic volume measured in vehicles per hour (vph) and vehicle counts. Traffic volume, measured by counting the number of vehicles N (reported by vehicle detectors) that pass a point on a road segment during time interval ∆t [113], is computed as v = N/∆t.

If placed upstream from the intersection, vehicle detectors can also estimate the number of stops that will occur given the current signal timing and historical vehicle counts (the historical vehicle counts are always from the previous five minutes from the request for an estimate). The time-space diagram, shown in Figure 4.3, illustrates how the number of stops are estimated (the same method is used in SCOOT). The number of vehicles detected (upstream) from point T_gc to point T_rc in time will reach the intersection during a red interval; these are the vehicles that will have to stop. Tgc

is the last time a vehicle can cross the detector during the green interval and make the light. And T_rcis the last time a vehicle will cross the detector and get stuck at the red light.

Figure 4.3: Time-space diagram for estimating stops. Vehicles that leave the up-stream intersection (labelled detector) between time T_gc and T_rc will reach the

down-stream intersection when the phase is showing red [1].