You can often tell how close it is to race day by the thickness of Madhur Behl’s facial hair. (Cavalier Racing photo)
University of Virginia engineering professor Madhur Behl believes the day will come, and the matchup will be historic when it does.
He imagines one of the world’s top racecar drivers risking his or her reputation - earned at speeds upwards of 250 mph and at G-forces exceeding what astronauts pull - to go axle-to-axle against at least one and maybe an entire field of autonomous racing competitors.
The human driver might even win the initial contest. But just as chess computers triumphed over grandmasters, Behl’s prediction is that artificial intelligence will ultimately prevail.
In fact, the 36-year-old sees that day as pivotal. That’s when he’ll know that self-driving cars are truly fit for our roads.
Behl, an associate professor in the departments of Computer Science, and Systems and Information Engineering, is pushing the limits of autonomous vehicle racing to make driving safer for humans.
Although he’s never physically steered anything more high-performance than his sporty Mazda MX-5 Miata, Behl and his students can say they’ve raced the Indianapolis Motor Speedway and other top tracks. UVA has competed against other collegiate teams in the Indy Autonomous Challenge since its inception in 2021.
Madhur Behl is an associate professor in the departments of Computer Science, and Systems and Information Engineering. (Photo by Dan Addison, University Communications)
Teams “race” by meticulously crafting precise computer code for their cars. In that respect, the $1.5 million in prize money may be a bargain for those handing it out. The data provided by each vehicle is expected to accelerate the commercialization of fully autonomous cars and advanced driver-assistance systems.
A photo of Behl’s original crew enshrines their place as pioneers of autonomous racing. In pit-style Cavalier windbreakers, the professor and his four students stand behind their sleek entry. The car’s body is identical to the other self-driving cars competing, with the exception of its paint job and sponsor logos; the car’s engine, as well as autonomy components such as sensors and the computer, are also identical.
Behl, center, stands flanked by his original pit crew members. From left, they are Jingyun Ning, Varundev Sukhil, Aron Harder and Amar Kulkarni. (Photo by Paul D’Andrea)
The differentiating factor lies in the team-designed algorithm and autonomous racing software responsible for the car’s maneuvering.
“Nobody in that picture set out to replace human racing,” Behl said. “That’s not what motivates us. We’re embracing the complexities of integrating AI technologies and robotics to advance safe autonomous driving.”
In the runup to a competition, the UVA team obsessively tests and tweaks, tests and tweaks.
“I take my entire Cavalier crew, and we go and live at the track for a few months,” Behl said. “The track becomes our lab.”
But this year will be a little different. Yes, the team is still heading to Indiana. But they won’t race the oval track there. Instead, they’ll be refitting the Hoo-mobile to race Formula One-style in Monza, Italy.
The Formula One Fan Tinkers
Behl, a Formula One fan since his youth, would be aghast if his research somehow made butt-in-seat racing irrelevant.
“My interest in racing was purely from a spectator’s point of view,” he said. “Of course, I would categorize Formula One as the most engineering-driven sport there is.”
A young Madhur Behl checks out a car while at a Madras International Circuit race in 2007. (Photo courtesy of Madhur Behl)
The son of a banker and a history teacher, Behl enjoyed a middle-class upbringing in Chandigarh, India, a city that he thinks of as relatively small, though it’s home to millions of people. He summered in the Himalayas when his father’s job moved him there seasonally. Behl’s two sisters went on to become doctors. He was the “outlier,” he said, interested in engineering.
As a student at Punjab Engineering College, Behl gravitated toward like-minded classmates. They started a robotics club. Their college was the only major engineering school that didn’t compete in his country’s prestigious “tech fest” competitions that posed annual challenges for students to solve. They sought to change that.
“We were wanting to be part of what was going on nationwide,” the professor said. “I have always built things, including a lot of robots. That was my main interest that got me into research and academia, and that’s where the seed of being competitive began.”
Behl recalled his first time traveling for a competition – two days by train to a fest at the Indian Institute of Technology Madras. As president of the club, he was his school’s lone representative, going up against about a hundred other teams, some of them “years ahead.”
The challenge was a robot that could climb a flight of stairs.
“We had this stair-climber that we built that was very bare bones,” he said. “It wasn’t like anything you’d imagine today with 3-D printed parts and polished electronics. It looked like something out of a junkyard. We had aluminum frames that we had bolted together, that we had to sometimes even weld together, to create a makeshift chassis on which the robot was designed. It had cams that allowed it to adhere to the profile of the stairs, and it climbed like a four-legged animal would.”
So how did the robot fare?
“We failed; we didn’t win anything.”
However, his team later won a competition with a robot that could kick a soccer ball.
Inventing F1/10 Racing
When Behl came to the U.S. to earn his master’s and doctorate in engineering at the University of Pennsylvania, neither robots nor racing were Behl’s official focus. Rather, his dissertation was on how to accurately predict and manage energy consumption at large scale – in a building, on a campus, even throughout a whole city.
He and his adviser, Rahul Mangharam, converted Behl’s algorithm into dashboard-style software that could recommend adjustments, taking into consideration the fluctuating cost of commercial electricity.
“We developed a tool that is used in many, many buildings today,” Behl said.
But as his more hands-on work wrapped up in 2015, the future UVA faculty member realized he still had some bliss he needed to follow. A master’s student in the lab where he worked, Paril Jain, shared his obsession with racing and automation. They challenged themselves to convert a 1/10 scale radio-controlled car into an autonomous vehicle.
They chose that particular-sized RC car for speed. “It’s very fast,” the professor said, motioning to one of the simple black racers with thick wheels. “It can go, like, 40 mph, and it’s impossible to catch if you run behind it.”
The pair spent several weekends adapting the car. Installing the software for the LIDAR system, which uses a laser to estimate the distance of objects, then integrating that software with the car’s controls, took up most of their time.
“So he helped me prototype this, and our ambition was just to amuse ourselves,” Behl said.
“I wanted to see if it could be done.”
This closeup shows off the features of the 1/10 size racer. (Photo courtesy of Madhur Behl)
Behl called the car type “F1/10,” a name combining the size of the car and his interest in Formula One. They ran the car around the lab and in the halls and corridors. And they posted information online as a guide for others.
“Within a year, it blew up, we got so much interest from our peer institutions,” he said. “I didn’t have the foresight to see this is where it was going to end up, but there was interest to use this as an educational tool for robotics for autonomous vehicles – to train a workforce.”
In retrospect, it might have been obvious. Accidentally crashing a cheaper, 1/10-size vehicle is way better than crashing a full-sized car.
The downsizing also allowed the cars to be subjected to indoor controlled experiments. Behl himself has 16 of the little cars in his lab in Rice Hall.
Hitting the Road for Research
With the growing interest in F1/10, Behl and colleagues gave tutorials at top academic conferences in robotics and cyber-physical systems, teaching others how to build their own F1/10 platform.
“Now people started saying, ‘I’m building three at my university. Next year, why don’t we come and we’ll have an autonomous race?’” he said.
“It’s truly mind-boggling. I think there’s something like 75 universities all over the world now that have used the car either for teaching or for research, or for both, or have come and competed in our official races. That’s thousands of people across several dozen really good universities all over the world.”
Those smart people are all collecting data through trial and error. They’re sharing what they find in research papers or in other, sometimes cooler ways – including making their original code available to others online.
Code, Behl said, is “where the thinking happens, and I often describe it as a recipe. Your ingredients are data from a LIDAR sensor, camera sensor, how far the wheels are turning and how fast the car is going. And so these are ingredients, you have to mix them, turn them and make sense of them, so that you can steer the car.”
He added, “Driving is really only two signals: how fast you move and how you steer.”
After formalizing the racing, Jain went on to work for Tesla’s Autopilot project, while Behl continued on the academic track. He joined the UVA faculty in 2017. Right away, he made F1/10 available to Virginia students through a for-credit course and club involvement.
So when the organizers of the Indy Autonomous Challenge began planning for their full-scale car competition, which would include university entrants, they naturally asked Behl for his advice.
He told them he would consult, but what he really wanted to do was race.
A Solid Track Record, Then a Crash
In the past several years, UVA teams have competed twice at Las Vegas Motor Speedway, once at Texas Motor Speedway and in the initial race at Indianapolis Motor Speedway.
UVA takes the spotlight at the first Indy Autonomous race, which was held at Indianapolis Motor Speedway. (Cavalier Racing photo)
They’ve experienced the thrill of making an actual-sized racecar whip around the oval at speeds up to 170 mph. In doing so, they’ve been able to claim the temporary bragging right of “fastest all-American self-driving race car team.”
But they’ve also experienced some tough breaks. Behl keeps a battered panel displayed on his office wall as a reminder of what can go terribly wrong.
“So far, we’ve had four races,” he said. “We did very well in half. We also had one of the worst crashes that the competition has seen. This is carbon fiber – or what happens to carbon fiber at 125 mph in a crash.”
Behl recently recounted for Automotive News’ “Shift” podcast how the car spun out at the CES consumer electronics trade show in Las Vegas last year while passing a competitor during a trial run.
“In racing there’s a saying: ‘If everything seems under control, you’re not going fast enough,’” he said.
The electronics, though, fortunately survived the accident unscathed.
Preparing for Monza
Now, the focus is on Monza. The legendary, winding circuit is the ultimate proving ground for autonomous vehicles.
At this year’s first “all hands” meeting, the professor and his returning team leads gave new members an idea of what to expect. (Photo by Dan Addison, University Communications)
That’s right, no more just turning left. Indy Autonomous will install what’s called a “slip differential” to allow the cars to turn right and make turns equally, without damaging their axles.
Though they won’t be responsible for their car’s mechanics, Behl warned the students in their first “all hands” meeting this summer to be prepared for some drastic curves – the learning kind.
“You will learn things that you never thought you would need to,” he said. “I’m a computer scientist. I never thought I needed to learn engine dynamics or how a racecar engine works or braking pressures, all of this stuff. But here we are now, we know everything about the suspension geometry and how the car has to be set up – the dynamics of everything.”
Among their tools, the students will be using three simulators, including one with a steering wheel and pedals. Their home base for code development will be the platform GitLab.
Being able to say you’ve solved problems on a self-driving vehicle obviously stands out on résumé. One student returning to the team, Shreepa Parthaje, recently landed an internship with autonomous drone company Skydio largely by showing race video and chatting up his role.
“Cavalier Autonomous Racing is a thrilling opportunity for students to get hands-on experiences in autonomous vehicle technology, something that is super-hard to come by,” Parthaje said. “As an undergraduate, I feel lucky to get to take part. Professor Behl’s guidance is something we value greatly, and tackling these problems provides practical skills and unique insights that directly translate to industry. Similar problems are being addressed by everyone right now, making this learning environment so awesome.”
Safety Born From Speed
Despite the fun he and his students are having, Behl is a serious researcher. He has won numerous awards in the areas of transportation, energy efficiency and robotics. The plaudits include taking the top honor at the World Embedded Software Contest, winning the Department of Energy’s “Cleantech” prize and, most recently, having been bestowed a CAREER award by the National Science Foundation.
He also shares his knowledge as an associate editor for the SAE International Journal of Connected and Automated Vehicles, guest editor for the Journal of Field Robotics, and as an academic advisory council member for Partners for Automated Vehicle Education.
He said racing is a logical place for his research. That’s where many of the major breakthroughs in automotive safety have been proved; seatbelts, rearview mirrors and advanced braking systems are all among the life-saving equipment that started out on the speedway.
“Almost every technology you can find in your sedan or SUV, you can track back to some innovation in motor sports,” he said.
But carmakers seem to have largely forgotten that tradition, Behl said. Those who have gone straight to the open road have done so with increased risks and uncertainties.
The academic’s prescription for how we get to safe self-driving – to essentially slow down and race – is actually two parts. For one, a large cluster of autonomous cars will need to one day compete without crashing into each other, or anything else. Then, two, an autonomous vehicle will need to win a race, again without catastrophe, against a pro.
If those conditions are met, Behl said, that’s when society can feel reasonably confident.
“Imagine the chaos of the Indy 500 with dozens of cars, all autonomous,” he said. “That would be a spectacle to be able to show the capabilities of this. So wouldn’t that be the car you would trust to navigate on a freeway? That’s where the bar has to be: if it can deal with the chaos of dozens of cars.”
He added that, while Indy Autonomous has the ability to race two cars simultaneously at high speeds, “So far, we’re not even close to what the pros can do.”
