Some of the greatest scientific achievements — digital photography, virtual reality, the Deep Web and the Internet — trace their origins to military research. The concept of self-driving cars too, owes a lot to the US wars in Iraq and Afghanistan. Competitions organised by Defense Advanced Research Projects Agency (DARPA) — a research and development wing of the Pentagon — in the mid-2000s have had a direct bearing on most of today’s autonomous vehicle (AV) projects.
The Stanford Cart, first built in 1961, could navigate around obstacles using cameras. Later, Carnegie Mellon researchers drove across the US in a NavLab 5, a Pontiac Trans Sport rigged up to drive itself using a windshield-mounted camera that looked for lane lines, while humans controlled the fuel pedal and brakes.
The first DARPA Grand Challenge of 2004 required robotic cars to travel 140 miles across the Mojave Desert. Carnegie Mellon University’s entry, a Humvee named ‘Sandstorm’, used cameras, laser scanners, radars, and a 1,000-pound box full of electronics to travel 7 miles. The final DARPA race in 2007, which was won by CMU’s ‘Boss’, followed by Stanford’s ‘Junior’ and Virginia Tech’s ‘Odin’, had a total budget of about $30 million.
Things have moved fast since then, with the military imperative being replaced by commercial considerations. The market for AV technology is valued at $100 billion, not including the value of the actual cars. Jefferies Research LLC recently valued Waymo, owned by Google’s parent Alphabet Inc., at $250 billion — over 8,000 times the 2007 DARPA budget.
In 2009, Google launched its self-driving car project with a team of DARPA Challenge veterans. A few years later, Tesla announced it would build a self-driving system into its cars. In 2015, Uber got scientists from CMU, a robotics and artificial intelligence powerhouse, for its project.
Two broad concepts are being tested:
- A system that uses radars, sonars, and cameras to perceive vehicles and other objects; this, according to a McKinsey report, requires less processing power, but does not assess the environment on a deeply granular level.
- The second approach uses Lidar — a remote sensing method that uses light in the form of a pulsed laser to measure variable distances and range — in addition to the traditional sensor suite of radar and camera systems. It requires more data-processing and computational power, but is more robust, especially in tight, traffic-heavy environments.
While the radar and camera technology is easy to optimise and robust enough to incorporate into mass-market cars, the challenge lies in leveraging artificial intelligence to convert 2D visuals into 3D images that the vehicle can then successfully negotiate. Lidar, on the other hand, is still expensive.
In an earnings call in February this year, Tesla founder Elon Musk dismissed the Lidar technology as being “too expensive” and “too bulky”, and defended Tesla’s strategy of achieving “full autonomy” using only cameras, radar, and ultrasonic sensors.
Lidar is a fixture on self-driving cars operated by GM, Uber and Waymo, the last two of which are currently fighting in court over Waymo’s allegation that Uber stole its Lidar technology. Tesla’s call to dump Lidar altogether is being seen as a new frontier.
The state of play
WAYMO launched Waymo One, a commercial self-driving car service and accompanying app for about 500 test families in suburban Phoenix, Arizona, in December 2018.
FORD Motor has set a date of 2021 for its first purpose-built driverless car. Ford plans to deploy “thousands of self-drivers” on the streets of multiple US cities in two years.
GENERAL MOTORS’ Cruise Automation plans to test a fleet of self-driving electric Chevy Bolts shortly. It has announced plans to debut its automated ride-hailing business in an American city (possibly New York) this year.
TESLA has claimed there will be self-driving Tesla “robotaxis” on the road as early as next year. Musk has said he would demonstrate a cross-country trip without touching the steering wheel.
VOLVO has labeled its AV project ‘IntelliSafe’, and set a zero-fatality goal before fully rolling out the autonomous features to the public. The Sweden-based carmaker plans to offer 100 Swedish customers early access to an autonomous XC90 SUV by 2021.
Riding on upbeat predictions, Lyft has raised $1 billion. Hundreds of smaller firms are rushing to offer better radars, cameras, Lidars, maps, and data management systems to the established players. Chipmakers such as Nvidia, Intel and Qualcomm are optimising power requirement by the cars, while Tesla has announced plans to make its own chips to meet specification requirements and cut costs.
Scepticism and concerns
After a 2018 incident in which an Uber self-driving car prototype fatally hit a pedestrian in Arizona, carmakers had paused to reflect on safety aspects. Uber, however, resumed testing its autonomous Volvos in December 2018, just nine months after the accident.
Ragunathan Rajkumar of CMU, who serves as co-director of the General Motors-Carnegie Mellon Connected and Autonomous Driving Collaborative Research Lab, is of the view that from a purely scientific standpoint, “Musk’s… proclamations would be laughed out of the room in academic circles and his submissions rejected summarily”.
Matthew Johnson-Roberson of the University of Michigan, who co-directs the UM Ford Center for Autonomous Vehicles, has said that the only way forward is to isolate autonomous cars in their own lanes, walled off from unpredictable humans.
On the Lidar-versus-camera debate begun by Musk, Rajkumar told The Indian Express that “the Tesla vehicles only have a camera and a radar, and they have been claiming for quite some time that the hardware will be capable of being fully autonomous with a future over-the-air update. Musk’s claim that Lidar companies are doomed is simply smoke and mirrors to confuse investors and consumers who are either not aware or not technology-savvy. No AV company today uses Lidars exclusively; they all have cameras too (and radars as well).”
In Rajkumar’s assessment, full autonomous driving “is some years away”.
A Deloitte study suggests that consumer trust in AVs is stalling. In the US, 50% of respondents did not believe they would be safe, nearly the same as last year’s 47%. The share of consumers in China, Japan and South Korea who believed AVs will not be safe decreased modestly; and Indian and German consumers both showed slight increase in distrust.