Four robot mini-cars. A complex network of tunnels. A race against the clock. This was the final test crafted by an MIT professor for what could be the greatest college class ever.

Despite the course title’s tortuous name — Rapid Autonomous Complex-Environment Competing Ackermann-steering Robot — its acronym, RACECAR, gets students’ curiosity revving.

The students’ challenge: Build a self-driving mini-robot car that can zip around a tunnel maze track while navigating its twists and turns.

Speed racer: MIT’s Sertac Karaman and his robot mini-car designed using the NVIDIA Jetson TK1 devkit.

To give the cars their autonomous abilities, students design and program algorithms using a Jetson TK1 embedded computer. Jetson TK1 helps the 1:10-scale cars deploy the open-source Robot Operating System, assess their environment and develop a language to help them race the fastest while careening around the course.

The course was so popular, its creator, Sertac Karaman, assistant professor of aeronautics and astronautics at MIT, is designing a bigger version for next year with Jetson at the core.

Raw Power

RACECAR students pool their talents, working in teams of five. Some have software skills. Others worked in robotics. A few have used robot operating systems. None had ever tried anything like this before.

While planning for RACECAR, unveiled last January, Karaman had a personal wish list.

“We needed raw computative power,” he said. “We needed cameras and sensors to feed a lot of data and process it very quickly. We needed powerful tools.”

After testing a Jetson, he ordered 10 more to help power six cars — four for students, one spare and one for he and his colleagues to work on.

Karaman, who joined MIT’s Aeronautics and Astronautics Department after completing his Ph.D. at the school, was struck by the robot-rich environment while learning about safety-critical embedded computer systems deployed in airplanes and drones. Having spent time working on autonomous cars and other vehicles, he was ready to add robots to the mix.

For RACECAR, Karaman and his collaborators — Michael Boulet, Owen Guldner and Michael Park from MIT’s Lincoln Laboratory Beaver Works Center — offered seven lectures on robot operating systems and how algorithmic robotic systems work, coupled with labs that included students learning to program a “scared car,” which backed up when someone walked toward it.

A three-day hackathon found the teams working flat out to implement software operating systems to prepare the cars to race through tunnels without crashing into things.

Robot Car Race

After some warm-ups along the race course, the competition began. The cars, designed not to exceed speeds of eight miles an hour even though they can reach more than double that, were raced one by one around more than 500 feet of tunnels.

“It took the first car about a minute and a half, but the team sped up, pushing the time down to just over 49 seconds. That’s an average speed of about seven miles an hour,” Karaman said. “The car was blasting around the corners. Even the rival teams cheered when the fastest team won.”

For next year’s course, Karaman has big plans — a Formula 1-style race arena with a dozen cars jostling for pole position. After harnessing the power of Jetson, he’s ready to add GPU-powered stereo cameras and feature detection.

“It’s surprising how fast people learn” with Jetson, he said. “You take it out the box and start playing with it. We’d like to put a Jetson on a drone that’s flying around. You can’t do that with a laptop.”

distributed by