Mercedes-Benz says it’s being forced to explore new ways of crunching huge amounts of data as it ramps up autonomous driving technology because its current processers are too slow and consume too much energy.
Existing Level 2 adaptive cruise control uses between 70-100W of energy, according to the German car-maker, but the more sophisticated Level 3 cruise, introduced on its latest S-Class, consumes around 400W.
The next-generation Level 4 driverless aids sucks-up as much as 3000W in operation while the final Level 5 being developed to see cars, vans and trucks operate without any human involvement at all will use as much as 20kW of power – and that’s not sustainable on its future electric vehicles.
This has forced engineers to look elsewhere for its solution to more efficient computing.
“The most efficient computer processing unit we know is the brain. As always, biology provides the answer, for a human to carry out the equivalent of L5 autonomy the brain consumes just 20W of power – a fraction of what we were achieving in the lab,” a Mercedes scientist told carsales.
Mercedes will eventually use new-generation neuromorphic computers – currently being developed – that are not only better and react faster are 10 times more efficient than current systems.
Sadly, to fully understand the tech needs a doctorate but the most basic explanation is that it uses silicon neurons to form a neural network that mimics the brain.
While existing tech wastes time processing everything a camera ‘sees’, the silicon neurons only react to spikes of information and the resulting event-triggered computation is both quicker and more efficient by up to 90 per cent in its current use, compared to existing tech.
