: To make an autonomous car safe, developers need to make it as smart as humans. At least, in certain aspects. To negotiate the unruly Delhi traffic, driverless cars need a whole different level of smarts. In Western cities, traffic rules are followed, signs are obeyed and right of way is given, not taken. It’s anything but in Delhi, and countless other cities in India.
That is why what Intel’s Corp.’s Mobileye unit has showcased through a 20-minute video of an autonomous car navigating its way through the streets of Jerusalem assumes importance. The unedited video showed the self-driving car making its way – the full video is available on its website — through road intersections without traffic lights, avoiding jaywalking pedestrians and swerving into oncoming traffic like an experienced Delhi cabbie. Like in Jerusalem, drivers in Delhi need to show an intent to move ahead or risk waiting endlessly for traffic to clear up.
Rowdy driving isn’t the only problem that Indian cities face. Barriers to the introduction of driverless cars are many in India – non-existent road markings, potholes and poorly maintained footpaths—but primary among them is the government’s unwillingness to allow it. In September, roads minister Nitin Gadkari ruled out the introduction of autonomous cars because it would leave thousands of people employed as drivers jobless.
But history shows that opposition, even from the government, has rarely, if ever, been able to stop the march of technology.
Mobileye, the self-driving technology company that Intel bought in 2017 for more than $15 billion, uses 12 cameras on its vehicle to navigate the roads. It doesn’t use any other sensors such as radars.
“Our goal is …to do everything that is necessary to do in autonomous driving without relying on any other sensor other than camera,” Amnon Shashua, chief executive of Mobileye, said at an event on Tuesday.
But the technology for autonomous cars is anything but perfect at the moment, according to Shashua who is also a computer science professor at Hebrew University in Jerusalem. Mobileye’s immediate target is to limit errors in car sensors to one in every 10,000 hours of driving, which matches with the probability of human-related failures in the US.
“We are targeting a failure in one after every 10,000 hours of driving. If we look at human statistics, 10,000 hours of driving is the probability of an accident with injuries. One million hours of driving, or around that number, is the probability of an accident with fatalities,” Sashua said at a CES 2020 event in Las Vegas. “But we would like a critical error in one in every 10 million hours of driving, which is an incredible number. There is no technology that is close to those numbers.”
Despite the challenges, Sashua expects to have full-stack autonomous driving technology for robo-taxis by 2022 and for consumers by 2025.
“The full stack self-driving system for consumers that we believe will come after robo taxi, by 2025 we will be able to price a self-driving system with end-to-end sensors, hardware, cable, everything below $5,000. That price point enables installing these systems into consumer vehicles. In the beginning, in premium vehicles but again this is where things start,” he said. “(It’s) the holy grail of this entire business, when you as a consumer can buy a passenger car that is autonomously enabled.”
The reporter is in Las Vegas at the invitation of Intel Corp.