traffic

As anyone who has driven on the UK's congested motorways will attest to, when the roads get beyond a critical threshold of overload, the sheer unpredictability of the drivers around you becomes the most important factor in your cognition. All it takes is for one driver to touch the brake pedal, lighting up the brake lights and a chain reaction of terror ensues in their wake. If an accident is luckily avoided then its almost a certainty that one of those frustratingly inexplicable causeless traffic jams will ensue. 

I have always believed in the power of computer network traffic engineering techniques to come to our aid in situations like this. Just like on a crowded pavement, the unpredictability of the individual has made such a solution frustratingly out of reach.

But it seems that automation and machine learning have brought this notion a step closer. By abdicating control to our machines, network traffic theory can be put into practice ensuring that optimal flow continues.

We have always lacked a way for vehicles to work together until recently and it is this collaborative effort overseen and perhaps controlled by a meta intelligence that can bring about the seismic change that has eluded us.

For my own part I detest most driving. Its basically dead time where my brain has to be used for this one mind numbing task despite the fact that I'd much rather be reading a book, getting some work done or even just sleeping. The day when I can tell my car where I want it to go and then switch off until Im there will be a red letter day for me. I was therefore recently pleased to hear the results of some recent research confirming that in tests, a fleet of driverless cars collaborating with each other can improve overall traffic flow by at least 35%.

Michael He, one of the researchers was quoted thus, "Autonomous cars could fix a lot of different problems associated with driving into, within and between cities but there has to be a way for them to work together."

The key will lie in the adoption of standards and, just like during the development of the standards which now dominate the internet, we are in a period of competition where the standard which wins out may not be the best. (Think ATM vs Ethernet for transporting video and VHS vs Betamax for watching it.)

Much of the current testing and development is done using scale models and SBC such as Raspberry Pi or Orange Pi. This enables researchers to avoid the prohibitive costs associated with developing full scale test environments. Using such swarm systems where the component nodes within the network are each able to communicate at least with their neighbours, it became possible for the overarching 'intelligence' to manage the meta priority for optimal traffic flow in such a way as to achieve something approaching harmony in a ballet of competing priorities and near misses that would send most human drivers to the hard shoulder. Cars can now be packed more closely and yet continue to enjoy progress towards the destination in environments which were previously untenable if populated by unpredictable humans.

Interestingly these tests involved simulating a mix of human and automata with the overall network collaboration level set to either egocentric or cooperative. Improvements of 35% were observed during cooperative traffic but during egocentric driving the improvement was as much as 45%.

Machine learning and swarm software modelling is bringing this field of imagined utopia into reality with staggering speed and for this driver, the day when I can tell my car where I'm going and then put my feet up can't come a moment too soon.