If it works here, it will work anywhere!
BANGALORE: Eight lanes of traffic are moving at snail’s pace. Bicycles, mopeds and pedestrians weave their way between the cars, buses and trucks.
No one pays much attention to the traffic lights in Bangalore, and when a cow ambles onto the road, everything comes to a halt.
Manu Saale looks out of his office window onto the chaos and shakes his head. As head of the Mercedes research centre in the Indian software capital, he faces a huge task — equipping the cars of tomorrow for the driverless era.
“The traffic is too complex to be compressed into algorithms that are able to foresee every scenario,” he says. “If at some point we want to be able to drive autonomously, the car has to take its own decisions, instead of following a programmed routine.”
The car then needs precisely what makes the difference between a good and a bad driver — experience. But exactly this is lacking in the current vehicle architecture.
Cars are now able to calculate extremely rapidly on the basis of dozens of control devices, but not think and certainly not learn, says Stefan Sommer, chief executive of automotive supplier ZF.
Sommer’s company has joined forces with chip manufacturer Nvidia to change the situation. “What the car needs for driverless motoring is AI [artificial intelligence],” Sommer says.
With the current mode of operation of vehicle electronics, in order to recognise a situation rapidly and call up the appropriate action, a driverless car has to interpret the results, make a decision and develop its own strategy, Sommer says.
This is much the way a learner driver operates. “With AI, the car collects experiences, as well as data, and is then better able to assess similar situations with each new experience,” he says.
The idea is training not programming, Nvidia head Jen-Hsun Huang says. Instead of having software for every conceivable situation, socalled machine learning, or deep learning, relies on the car recognising an obstacle and knowing when to avoid it or stop.
“The software must patterns and draw conclusions,” he says.
Sommer’s company aims to have their controlling device with Nvidia chip ready for series production within a few years. “With it we will apply the computing power of a supercomputer, as it is needed for AI, on the road and allow cars not only to see, but also think and act,” he says. But not everyone is convinced. Nissan and Renault head Carlos Ghosn is not ready to allow the computer to take over, believing that the human driver needs to be ultimately in charge.
For driverless Nissan recognise the right models he has joined forces with NASA to install a kind of “Mission Control” where human experts will resolve problems by remote control, as Melissa Cefkin of Nissan’s Silicon Valley research centre says.
But they too aim to make the software intelligent to the extent that human intervention is rarely needed.
Manu Saale knows that it will be years before cars will be able to learn and build up a store of experience that they can use to make decisions in the chaos he observes from his office window.
But he remains optimistic, and of one thing he is certain: “If we manage it on the streets of Bangalore, the autopilot will work anywhere.”