One quarter of all accidents and a third of all fatal accidents occur at intersections. The accidents themselves cause loss of life, and the resulting traffic problems cause loss of productivity and an increase in traffic pollution. Traffic lights are one of the ways in which we control traffic flow, but they are not as efficient as the could be.
Stopping and then accelerating at intersections is an unnecessary cause of pollution and carbon emissions, and it also delays drivers which adds economic cost to businesses that rely on freight and deliveries.
Research into traffic light phasing and autonomous (driverless) vehicles by Massachusetts Institute of Technology (MIT) Senseable City Lab, the Swiss Institute of Technology (ETHZ) and the Italian National Research Council (CNR) has resulted in a slot-based system of allowing vehicles just enough time to pass through an intersection before another vehicle comes. Their model cuts delays in half, doubling the traffic throughput (however, it doesn’t account for bottlenecks that occur in other areas that could hold traffic up at the intersection).
This was already proposed in 2012 by Peter Stone, Professor of Computer Science at the University of Texas.
As traffic congestion costs exceed US$5billion per year in the USA alone, solving this problem has potentially huge payoffs. The challenge will be to secure the communications between vehicles and the intersection. How vulnerable will the artificial intelligence systems be to denial-of-service attacks, spoofing and other hacks?
Of course, all vehicles must be autonomous for this to work effectively and that could be decades away given that the average age of our vehicle fleet is over 14, and the average age at which they’re scrapped is 18 years. We have the technology right now to implement this but the costs are prohibitive both for transport agencies and the vehicle-owning public.
What will the autonomous vehicle system and intersection artificial intelligence need to know?
Let’s assume every vehicle on the road is autonomous. In order to function properly each vehicle approaching the intersection, as well as the intersection’s artificial intelligence, will need to know:
- How fast is the vehicle travelling
- In what lane is the vehicle travelling
- Where is the vehicle going, e.g. straight through, turning left, etc
- How long is the vehicle (this affects the turning time)
- What type of vehicle is it (heavy vehicles can’t turn as fast as motorcycles and cars, and is it possible to have autonomous motorcycles)
- Are there any emergency scenarios e.g. is an ambulance or police car coming
- What is the current weather (weather affects grip on the road, and icy conditions could mean slower turning times)
- Is this a genuine vehicle, i.e. not a spoof or other hack
- Are any of the vehicles experiencing mechanical difficulties
- Has an unexpected object entered the intersection
- Do any of the vehicles detect an unexpected object
One factor that is not outlined in the videos is pedestrians and cyclists. Autonomous vehicles will have the capability to avoid collisions with them but as they can be erratic and unpredictable, the slot that remains open for the vehicle through the intersection could close quickly. This would mean that vehicle speeds would still have to be reduced to enable them to stop or take evasive action in time.
In some scenarios a roundabout is a much better solution. However, roundabouts in high-traffic scenarios can still cause bottlenecks.