Teijin begins operations at carbon fiber reinforced thermoplastic plant; targeting components for mass-produced vehicles
UMTRI: November average new vehicle fuel economy in US same as October; September Eco-Driving Index at a record 0.8

VTTI team proposes optimization algorithm for driverless vehicles at unsignaled intersections

The layout of the proposed multi-agent system (MAS) for driverless vehicles at intersections. Zohdy and Rakha. Click to enlarge.

Researchers at Virginia Tech Transportation Institute (VTTI) have developed a heuristic optimization algorithm for driverless vehicles at unsignalized intersections using a multi-agent system (MAS). Their research, presented at the Intelligent Transportation Society World Congress in Vienna in October, won the Best Scientific Paper Award for North America.

The system proposed by Ismail Zohdy, a Ph.D. student in civil engineering at Virginia Tech, and Hesham Rakha, director of the Center for Sustainable Mobility at the transportation institute and professor of civil engineering at the university, models the driverless vehicles as autonomous agents controlled by the intersection controller (manager agent).

...it is anticipated in the future that many (or most) of the vehicles will be fully automated; thus the movements of those vehicles will need to be optimized in the network. Imagine that all running vehicles are unmanned and controlled by highly sophisticated equipment, there will be a need for innovative optimization algorithms for controlling these driverless vehicles. This research effort attempts to focus on optimizing the movements of the future intelligent (driverless/autonomous/unmanned) vehicles at unsignalized intersections by controlling these vehicles as agents that have certain goals and limitations.

—Zohdy and Rakha 2012

Input information consists of vehicles’ current location, speed and acceleration in addition to the surrounding environment (weather, intersection characteristics, etc.). The intersection controller processes the input information using a built-in simulator: “OSDI” (Optimization Simulator for Driverless vehicles at Intersections).

The objective of the simulator is to optimize the movements of vehicles to reduce the total delay time for the entire intersection and prevent crashes simultaneously. The intersection controller then uses the simulator output for controlling the speed profile of the driverless vehicles within the intersection study zone. In Zohdy and Rakha’s research, the intersection controller governs the vehicles within 200 meters (218.7 yards) from the intersection.

In the paper, the team compared the proposed system to a control scenario (all-way stop control, AWSC) assuming that there are four driverless vehicles (one vehicle per approach) willing to cross a four-legged unsignalized intersection concurrently. Monte Carlo simulation results show that the proposed system reduces the total delay by 35 seconds on average compared to traditional AWSC.

The research has since been expanded and tested on more congested intersections involving not fully deployed systems and comparing this type of control to traffic signal and roundabout control.

We were testing it if only 10 percent of the vehicles were automated and the other 90 percent were regular vehicles with driver control. We varied the level of automation from 10 to 100 percent at 10 percent increments.

—Hesham Rakha

This effort has resulted in two papers that will be presented that the Transportation Research Board Annual Meeting in January 2013.

Zohdy and Rakha will also be testing their system on a roundabout on the Virginia Tech campus as part of the Connected Vehicle/Infrastructure University Transportation Center.




This is Key - how does an automated car deal with heavy, greedy traffic, as well as multi-automats at a junction.

Do you have to automate rule breaking so that an automat can push out into heavy after waiting for a certain amount of time ?

How will the rules vary as the % of automats increases from 0 - X % automatic cars?


Passenger planes are good examples to follow. They currently avoid near misses and collisions. Future smart vehicles will be able to navigate in and out of heavy traffic without accidents, specially when the majority will be equipped with driver less systems.

Human drivers are responsible for 80% to 85% of all accidents and road fatalities. The solution is to assist them to do better and eventually to replace them.


These kinds of cars will be viable only if passangers are required to wear blindfolds. I predicted years ago that automatically driven cars will soon be whizzing through intersections without traffic lights, at high speeds, through the smallest gaps between cars moving through on the crossing road. Imagine the increase in heart attacks if people have to watch this kind of controlled chaos from inside their cars. You think calculating, prize-winning engineers think about that? No!

I don't want human factor-disconnected engineers determining my chance of survival based on their ability to write an algorythm. I trust my own ability to spot red-light runners by paying attention, looking both ways before going through any green light, and slowing down when cell phone talkers are behind me.

By the way, why can't these prize-winning engineers at Virginia Tech (where I got my EE degree) invent a traffic light that doesn't change, when a car makes a right turn on red?

The comments to this entry are closed.