The great robot race : A turning point in the quest for Autonomous Driving

The beginning couldn't have been worse for such an ambitious initiative. In 2004, the reputed US Agency DARPA (Defense Advanced Research Project Agency), announced a nationwide competition on autonomous driving. The competition was tough, it required an autonomous vehicle to successfully negotiate the rugged terrain of the Mojave dessert in the US and cover almost 150 miles which included off-road paths. There was an enthusiastic response from close to 200 participants, out of which around 50 were shortlisted to participate in the qualifying round.

The end was totally anti-climactic. None of the 20 finalists teams could complete the race. The maximum distance covered by a driverless vehicle was only 7 miles. Most vehicles crashed and burned in the challenging terrain, after spinning their wheels endlessly around numerous obstacles which showed up everywhere. Clearly the robotic enthusiasts of the time underestimated the complexity of the terrain and overestimated the adaptive capabilities of their programs.

No team won, and only one team was given a prize, as a consolation for efforts.

The 2005 competition

DARPA, to their credit, took the failure in their stride and announced the next round of competition in 2005. During the qualifying rounds, the participating vehicles were required to demonstrate their ability to navigate a variety of terrain and obstacles, including paved and unpaved roads, tunnels, and bridges.

The qualifying event was a kind of Woodstock for robot geeks. The speedway was packed wonderful engineers cheering their magnificent driving machines. The first to go was a self-driving motorcycle, very aptly named as "Ghostrider" who promptly hit the obstacle and tumbled, only to rise again and continue its ride. Overall, the atmosphere was electric, almost akin to excited parents watching their kids perform in an arena, cheering and reacting to each and every step.

A total of 23 teams qualified for the final round of the 2005 competition, representing a variety of institutions including universities, private companies, and government agencies. The goal of final round was to complete the course as quickly as possible without human intervention.

The results

Unlike the 2004 competition, in which none of the vehicles were able to complete the course, 5 teams were able to make it to the finish line in 2005. The winning team, Stanford University's Racing team, known as "Stanley", completed the course in a time of 6 hours and 53 minutes, with an average speed of 19.1 mph and won the $2 million prize. What makes their win even more remarkable, is that they were participating in the competition for the first time.

The Stanford Racing Team's autonomous vehicle for the 2005 DARPA Grand Challenge was a re-engineered Volkswagen Tuareg R5. The Stanley team treated used a combination of various sensors to understand the terrain :

• Radars to capture long distance inputs. They could "see" up to 200 meters but information was sparser.
• LIDARs to capture short range inputs up to 25 meters which was of higher fidelity. In the picture, 5 LIDAR sensors can be seen mounted on the top of the vehicle.
• Video based inputs to capture everything in between

The Stanley team treated the challenge as a software problem and designed a multi-stack software to read the terrain and navigate the vehicle. The software stack consisted of three modules, which included a Data Acquisition module which would process all sensor inputs, a planning module which was trained using Machine Learning to plot the course and finally a module to execute the driving plan, in terms of speed and driving response. The controller was written in hundred thousand lines of code, was deployed in a 6 core Pentium computer in the back of the car.

Carnegie Mellon University narrowly missed winning the race. It had fielded two teams, "Highlander" and "Sandstorm" which successfully finished the race in second and third positions. In fact, Highlander had a great start and seemed to be the favorite to win the race, but some unforeseen technical problem slowed it down and it was overtaken by Stanley during the course.

Impact and Legacy

The 2005 DARPA Grand Challenge was a significant milestone in the development of autonomous vehicles, as it marked the first time that a self-driving vehicle could successfully complete a course of this distance and complexity. It demonstrated the feasibility of driverless cars in an extremely rugged terrain and the capabilities of the hardware and software needed to make it happen.

Members of the Stanley team later joined Google, where they spearheaded its Waymo project for the driverless car.

It helped to pave the way for further advancements in the field and contributed to the development of self-driving cars in the years that followed. The LIDAR vs Camera debate continues till today.

Now in 2022, autonomous driving is already in operation in select locations. It holds great promise to help with road safety and with reducing accidents due to human errors, ensuring better utilization of roads as self driving cars can be positioned closer to each other due to their faster reaction times, and for helping people who are unable to drive. The race sponsored by DARPA was a watershed moment in this quest.

PS : Thanks for reading ! The views expressed here are personal.