City Brain Challenge

In this challenge, we will provide you with a city-scale road network and its traffic demand derived from real traffic data. You will be in charge of coordinating the traffic signals to maximize number of vehicles served while maintaining an acceptable delay. We will increase the traffic demand and see whether your coordination model can still survive.

Problem

Traffic signals coordinate the traffic movements at the intersection and a smart traffic signal coordination algorithm is the key to transportation efficiency. For a four-leg intersection (see figure below), 1 of 8 types of signal phases can be selected each period of time step, serving a pair of non-conflict traffic movements (e.g., phase-1 gives right-of-way for left-turn traffic from northern and southern approaches). In this competition, participants need to develop a model to select traffic signal phases at intersections of a road network to improve road network traffic performance.

https://raw.githubusercontent.com/CityBrainChallenge/KDDCup2021-CityBrainChallenge/main/images/intersection.png

In the final phase, a city-scale road network sample traffic data is provided. We use exactly the same road network but different traffic data for scoring your submissions. Participants are encouraged to use the python script to generate your own sample traffic data for training and testing since the traffic settings for evaluation is not revealed.

Evaluation

Total number of vehicles served (i.e., total number of vehicles entering the network) and delay index will be computed every 20 seconds to evaluate your submissions. The evaluation process will be terminated once the delay index reaches the predefined threshold 1.40.

The trip delay index is computed as actual travel time divided by travel time at free-flow speed. For an uncompleted trip, the free-flow speed is used to estimate the travel time of rest of the trip. The delay index is computed as average trip delay index over all vehicles served: \(D = \frac{1}{N}\sum_{i=1}^{N}{D_{i}}\).

The trip delay \(D_{i}\) of vehicle \(i\) is defined as \(D_{i} = \frac{TT_{i} + TT_{i}^{r}}{TT_{i}^{f}}\), where,

  • \(TT_i\): travel time of vehicle \(i\);

  • \(TT_{i}^{r}\): rest of trip travel time, estimated with free-flow speed;

  • \(TT_{i}^{f}\): full trip travel time at free-flow speed

We will evaluate your solution on multiple traffic flow settings. The total number of served vehicles is computed over all evaluation scenarios. The overall delay index is computed as the average delay index among all vehicles of all scenarios.

The submission scoring and ranking process follows three principles:

  • Solutions that served more vehicles will rank higher.

  • If two solutions served the same number of vehicles, the one with lower delay index will rank higher.

  • If two solutions served the same number of vehicles with same delay index, the one submitted earlier will rank higher.