Avoiding gridlock using cognitive architecture and smart traffic management systems

Cognitive architectures are frameworks that attempt to model how cognition happens, in the brains of humans or animals. Implemented in software, such architectures can be used to develop intelligent agents capable of controlling and managing cyber-physical systems. In a recent collaboration project between Ericsson Research and the University of Campinas, we have applied cognitive architecture technology to the control and management of urban traffic lights in a simulated smart city scenario.

Street Macro 6701 in enviroment

Ever-increasing traffic in big cities constantly creates huge traffic jams. Current approaches to traffic light control have not been able to properly avoid traffic overload. But maybe cognitive architecture technology could be used to overcome this problem?

There are many cognitive architectures implemented in software, each with their own specific proposals on how to model cognition for implementing intelligent agents. However, many of them are not easy to use and sometimes hard to integrate with existing systems. To address these issues, our team of researchers from Ericsson Research and the University of Campinas developed MECA – Multipurpose Enhanced Cognitive Architecture.

Then we went on to apply this architecture to the problem of traffic control. We created a system composed of a set of cognitive managers, that is, control agents with cognitive capabilities, each being responsible for controlling the traffic lights of a single junction in a city topology as can be seen in the following diagram.

Cognitive Architecture 1


What we have above is, therefore, a community of intelligent agents, or cognitive managers, each taking care of its own traffic junction but able to exchange information with neighbors in order to improve traffic in a global manner.

A more detailed view of a traffic junction can be seen in the following figure. There you can see a "simple-T" type of intersection. Traffic can flow in many directions according to the state, or phase, of traffic lights in the junction. Each phase is a valid configuration that can be chosen by the cognitive manager taking care of this junction, i.e. only safe and feasible phase configurations are available for choosing. To make a decision, the cognitive manager takes into consideration information from induction loop sensors on the ground, for estimating traffic at the junction, and information from other cognitive managers in the network.

Cognitive Architecture 2


Each cognitive manager is composed of the elements seen in the next figure. The mind of the intelligent agent is composed of two sub-systems: system 1, which is faster and more reactive, and system 2, which is slower and more deliberative. System 1 is ideal fast closed-loop control, while system 2 focuses on solving unexpected or special situations. System 2 must influence system 1, when dealing with those situations becomes necessary.

Cognitive Architecture 3


Experiments were done in different simulated urban traffic network topologies using the SUMO (Simulator of Urban Mobility) simulator. The SUMO simulator is a software tool developed by the Institute of Transportation Systems at the German Aerospace Center (DLR). It is an open source microscopic and continuous road traffic simulation platform and is widely used in urban traffic research, which is why we have chosen this specific simulator for our experiments.

Below are examples of traffic simulation scenarios widely used in research.

Cognitive Architecture 4


The SimpleT scenario is the simplest, with just one traffic light, considered in the middle of a road where a T-like shaped sub-road creates an interruption in the traffic.

The TwinT scenario is a small variation of the SimpleT scenario, where the major road is interrupted by two consecutive T-shaped intersections. This scenario has 2 traffic lights and allows the study of the effect of one traffic light in its neighbor.

The Corridor scenario introduces a further complication, with 4 traffic lights and with different kinds of intersections.

Finally, the last scenario that was studied was the Manhattan scenario, named after the typical configuration encountered in Manhattan, New York City. With a grid-like topology it introduces further complications such as allowing cars to circulate in loops and performing more complex trajectories.

Results show that the system is able to provide a fair service, providing collaboration with Smart Cars (cars with special priorities such as ambulances and fire fighters) when necessary.

Cognitive Architecture 5


In the above figure, you see three sets of simulation results for the Manhattan scenario. Red curves are for the fixed traffic lights and the green ones are for traffic lights controlled by cognitive managers. The bold line is the average of 10 different simulations with the same traffic densities. The light colors extend from the minimum up to the maximum values over these 10 simulations. It compares average waiting time when applying fixed transitions (red) with cognitive managers with only system 1 active (green). A remarkable result is given by the graphics in the middle. The use of cognitive managers avoided a major traffic jam which happened in most cases with the fixed controllers.

We analyzed the performance of Smart Cars over normal cars for the Manhattan scenario, in three different traffic densities. The below figures illustrate the results.

Cognitive Architecture 6


We ran 10 different simulations for each traffic density. Then, for each simulation, we chose the car with the worst waiting time, promoted it to a Smart Car and re-ran the same simulation, now using System 2 to give priority to Smart Cars. The bars in red show the waiting time for these cars when only System 1 was running, and the blue bars show the waiting time for these cars when both System 1 and System 2 were running.

It is our goal to make MECA as general as possible, so it can be applied to a variety of domains. As one step towards this, our group is now working on a project applying MECA to a smart manufacturing scenario, more specifically warehouse automation with mobile robots.

Learn more about MECA at its github repository: https://github.com/EricssonResearch/meca

Learn more about Cognitive Architecture technology at: https://www.ericsson.com/research-blog/machines-partners-rather-replacements/

The bulk of results discussed in this blog post were published in October 2018 in the Biologically Inspired Cognitive Architectures volume 26 article "An Urban Traffic Controller Using the MECA Cognitive Architecture" by Gudwin, Ricardo, André Paraense, Suelen M. de Paula, Eduardo Fróes, Wandemberg Gibaut, Elisa Castro, Vera Figueiredo, and Klaus Raizer.


The Ericsson Blog

Like what you’re reading? Please sign up for email updates on your favorite topics.

Subscribe now

At the Ericsson Blog, we provide insight to make complex ideas on technology, innovation and business simple.