How AI orchestrates 4 areas central to autonomous cars, Auto News, ET Auto

The car is becoming fully automated, and transforming into an experience zone. In becoming so, a lot of investment is going into four areas, and AI is the key technology orchestrating all of those areas, Anurag Sahay, MD & global lead for AI & data sciences at digital product engineering services company Nagarro, said at our webinar last week.

The first area, he said, is localisation and mapping – getting the car to know where it is exactly, which lane it is in, how fast it is going. The second is scene understanding – getting the car to understand what’s happening around it, map every other car, map pedestrians, their movements, project where they will be five seconds later. The third is to plan a journey – through highways, intersections, traffic lights – and dynamically change it if necessary. The fourth is to make the car understand the internal state, the people sitting inside, are they ok, is there a health problem emerging, is the AC fine.

Sahay said large data sets are needed to train the AI algorithms. Tesla for instance, he said, has 1.5 petabytes of training data under different conditions, and even that is not enough. And so it’s continuously collecting more from its cars. Then you have to label all that data, for which you need extraordinary tools. And finally, you have to train the AI model. “To train a model that can bring a self-driving experience, it takes 10,000 GPUs for two weeks. That involves enormous cost, and mind boggling distributed engineering,” he said.

Rishi Raj Ranga, director of business solutions at Nagarro, said introducing a new mobility use case or solution is one thing, but to tailor it to a city’s construct is a different thing. “You have to make sure all your mobility assets are optimally utilised, that mobility systems are responsive enough, they are efficient. There are a lot of innovations possible, and I’m betting that data will be the oxygen that this entire mobility operating system needs in order to be sustainable, efficient and responsive to consumers’ needs,” he said.

Sven Sommerfeld, MD of ATCS, a tech consulting company that Nagarro acquired last year, said traditional automotive companies will have to become hugely agile and a lot more technical to be able to do all the complex engineering required and deliver to customers the continuous value they expect today. “You also need to do deep integration from the chip to the software to the car. Not everyone can do that. So you need partners who come in with an agile and product-oriented mindset,” he said.

Also Read:

The car is becoming fully automated, and transforming into an experience zone. In becoming so, a lot of investment is going into four areas, and AI is the key technology orchestrating all of those areas, Anurag Sahay, MD & global lead for AI & data sciences at digital product engineering services company Nagarro, said at our webinar last week.

The first area, he said, is localisation and mapping – getting the car to know where it is exactly, which lane it is in, how fast it is going. The second is scene understanding – getting the car to understand what’s happening around it, map every other car, map pedestrians, their movements, project where they will be five seconds later. The third is to plan a journey – through highways, intersections, traffic lights – and dynamically change it if necessary. The fourth is to make the car understand the internal state, the people sitting inside, are they ok, is there a health problem emerging, is the AC fine.

Sahay said large data sets are needed to train the AI algorithms. Tesla for instance, he said, has 1.5 petabytes of training data under different conditions, and even that is not enough. And so it’s continuously collecting more from its cars. Then you have to label all that data, for which you need extraordinary tools. And finally, you have to train the AI model. “To train a model that can bring a self-driving experience, it takes 10,000 GPUs for two weeks. That involves enormous cost, and mind boggling distributed engineering,” he said.

Rishi Raj Ranga, director of business solutions at Nagarro, said introducing a new mobility use case or solution is one thing, but to tailor it to a city’s construct is a different thing. “You have to make sure all your mobility assets are optimally utilised, that mobility systems are responsive enough, they are efficient. There are a lot of innovations possible, and I’m betting that data will be the oxygen that this entire mobility operating system needs in order to be sustainable, efficient and responsive to consumers’ needs,” he said.

Sven Sommerfeld, MD of ATCS, a tech consulting company that Nagarro acquired last year, said traditional automotive companies will have to become hugely agile and a lot more technical to be able to do all the complex engineering required and deliver to customers the continuous value they expect today. “You also need to do deep integration from the chip to the software to the car. Not everyone can do that. So you need partners who come in with an agile and product-oriented mindset,” he said.

Also Read: