The sport of racing is a balance of strategy and improvisation, requiring intense focus to make split-second decisions. But when there’s no human pushing the pedal to the metal, how do you break away?
Automotive enthusiasts will have a chance to see the answer in mid-July, when Robocar, the world’s first robotic race car, attempts the first autonomous hillclimb at the 25th anniversary of the Goodwood Festival of Speed.
The demonstration is just the latest step as researchers and entrepreneurs work to to turn autonomous racing into a spectator sport. In May, a team of researchers at the Technical University of Munich (TUM) using the NVIDIA DRIVE PX 2 autonomous driving platform built into the Robocar won the first Roborace Human + Machine Challenge in Berlin.
“NVIDIA DRIVE is the only hardware setup which can process these sensors so fast,” said Johannes Betz, a researcher at TUM’s automotive technology program. “You could try to use a regular computer, but it’s not suitable for use in vehicles.”
Racing: the Ultimate Test for DRIVE PX 2
Call it the ultimate test for our DRIVE platform. Roborace is a new type of racing series designed for the autonomous driving era that uses a specialized vehicle built without a driver’s seat.
Run by a team of venture-backed entrepreneurs, Roborace has already already become a sensation among automotive enthusiasts. It doesn’t hurt that automotive futurist Daniel Simon — known for his work in Hollywood films such as Oblivion and Tron: Legacy — designed the sleek, futuristic autonomous racers.
Under the hood, four 135kW electric motors power each wheel for a combined 500-plus horsepower. An NVIDIA DRIVE computer processes Robocar’s data, which includes inputs from the lidar, radar, GPS, ultrasonic, and camera sensors.
The Roborace platform also includes access to their DevBot vehicles, which have the same technical set up as the Robocar, but allow for both human and autonomous driving. To make it to the checkered flag, teams must develop algorithms capable of perceiving a vehicle’s environment, calculating speed and controlling the car with precision.
TUM competed against the University of Pisa in Roborace’s Human + Machine Challenge, the first prototype competition featuring two teams using the company’s DevBots. Teams had to drive the DevBot cars three times around the 2-kilometer track with a human at the wheel, then run the vehicle three more times on software alone. The timing of the second lap of each round is averaged, and the team with the fastest average wins.
Developing in the Fast Lane
Programming an autonomous racecar is no small feat, especially when a team only has a matter of months to complete development. The TUM group got to work on the challenge in January after establishing its team of seven core researchers from two of the University’s institutes, the Chair of Automotive Technology led by Markus Lienkamp and the Chair of Automatic Control headed by Boris Lohmann.
The team divided the software development into perception, strategy, planning and control, ensuring the algorithms could handle all driving tasks autonomously. To test the software, the researchers traveled to Roborace’s headquarters in Banbury, Oxfordshire, each month to run the algorithms on a working DevBot, also powered by the NVIDIA DRIVE PX 2 platform.
The development process requires just as much computing horsepower as the final race. The TUM team used the NVIDIA AI platform in their own lab, which allowed them to run software in real time on an embedded platform for automotive for split-second decision-making and sensor fusion.
The five months of development and testing on the durable processing platform paid off, with the TUM DevBot finishing the Human + Machine Challenge with an average lap time of 91.59 seconds, nearly four seconds faster than the University Pisa’s time of 95.36 seconds.
From the Track to the Streets
The victory is just the beginning of what Betz sees as an ongoing project for the TUM team. Autonomous racing allows researchers to take self-driving software and hardware to its limits — faster speeds, tighter turns and quicker path planning — and validate the technology quickly on repeated, closed course loops.
Betz said he wants to continue to compete in Roborace events to refine the software, educate new autonomous vehicle engineers and research ways to apply the technology in the real world.
“That’s the reason why we decided to take part in this,” Betz said. “Everything we are developing can be leveraged and brought back to a normal street vehicle.”
Racing has long been a testbed for new automotive technologies, from the seatbelt to the rearview mirror. Now, Betz and his team see it as the next frontier to test autonomous capabilities such as perception, object detection and path planning.
“We see it as a long-term project,” Betz said. However, he added, the team still has a competitive edge.
“We want to win the races.”
To learn more about Roborace and TUM, check out NVIDIA GTC Europe, where the two groups will host a session.