Sometimes, you have to embrace the challenge.
Sometimes, if you’re Lantao Liu, Associate Professor of Intelligent Systems Engineering and Computer Science, and his Luddy School of Informatics, Computing, and Engineering’s Vehicle Autonomy Intelligence Lab team, you beat the odds as well as more experienced autonomous racing squads from around the world in an impressive debut in the Indy Autonomous Challenge.
After a strong showing in the Challenge’s simulated-race phase, and with real-world autonomous racing next, this much is certain:
The best is still ahead.
“We want to make something cool,” says Liu, who is also the VAIL director. “Our lab, and the entire university, will benefit with the visibility. It’s possible that in two years we can climb up to the level of the long-term players from the European countries.”
VAIL focuses on developing methodologies that enhance the autonomy and intelligence of robotic systems such as unmanned ground, aerial and aquatic vehicles. Its research ranges from drones to robotic cars to even robotic dogs that walk, dance, roll over and play dead and rear up as if begging for food.
The Luddy team is developing the software for an AI-controlled racing car. The objective is to build a robust code to handle any racing situation.
Prospects are promising after a successful series of simulated races. The Indy Autonomous Challenge culminates in a June race in Milan, Italy, using 1,400-pound racing cars that can reach speeds topping 186 mph.
Teams use an Indy Autonomous Challenge-built AV-24 that’s been retrofitted with software, hardware and controls to allow autonomous driving. The AV-24 chassis is a modified version of the Indy Lights chassis. It’s a collaboration of Dallara IndyCar Factory in Speedway, Indiana, and Dallara’s Italian headquarters in Italy.
For now, the VAIL team is testing with small robotic cars. It is assembling a go-Kart to test. It expects to get a real car by the first week in April. That will give the team around two months before the Milan race.
VAIL’s previous work with AI and autonomous vehicles have helped the team quickly close the Challenge gap.
“Here we do autonomous navigation for small robots,” Ph.D. student Mahmoud Ali says. “That navigation is similar for small and big robots. We have already trained on the autonomous track. The biggest challenge is going from low to high speed.”
The Luddy team was assembled last August with consultants from an industrial partner, Code19, making it a rookie facing veteran squads with years of experience. Adding to the challenge -- the Luddy team had to be ready for a three-phase series of simulated races that began in September and ended in December.
“The timeline was very short,” Liu says, “but we have the expertise for this.”
To facilitate the process, Liu assembled eight sub-teams each led by a Ph.D. student who recruited a couple of undergraduates to help.
The technical leads are Ali, Durgakant Pushp, Ihab Mohamed, Hassan Jardali, Paul Coen, Youwei Yu, Md. Al-Masrur Khan and Alejandro Murillo Gonzalez. Abe Leininger is an undergraduate lead.
Liu says 150 students applied and 28 people are on the project. Teams worked 20 or more hours per week for six straight weeks to get ready for the simulated races.
The Luddy team has thrived with ingenuity, innovation, effort, strong research and plenty of help from the Luddy School and Dean Joanna Millunchick.
“Luddy is very supportive,” Liu says. “Dean Millunchick has done her best to provide us with the resources, collaborators and many other things we need. That speeds up our preparation from many perspectives, including resources and hardware. I appreciate that.”
In the first simulated race, basically a timed qualifier with cars racing by themselves, IU placed fourth among 18 teams from around the world, and was first among U.S. squads, with a one-lap time of 118.75 seconds.
For the second race, which involved competing against AI opponents, the IU team developed an overtaking algorithm to pass other cars. The team passed four cars before crashing on another passing attempt and finished eighth.
In the final simulated race, organizers provided erroneous sensor data that teams had to overcome while avoiding crashes and improving passing skills. Despite a track violation, the IU team passed five AI opponents and cut its minimum lap time to 115.42 seconds. That was good for sixth.
“I’m happy with the result,” Liu says. “There was uncertainty, but we made it happen. The ranking was good given we only had three months of preparation and the other teams were long-term players.”
All this was done under simulated conditions. Adapting to real-world racing adds significant complexity. Liu calls it the “simulation to real-world gap.”
Speed, control, awareness and even weather are factors. Drivers -- human or AI – make non-stop multiple decisions under conditions that can change in an instant.
“Going from simulation to actual race, that challenge will be huge,” Liu says. “Put HUGE in capital letters, especially because we don’t have the car yet.
“Many times, everything works well in a simulation, but you try it in the real world with a real car and many components, maybe none of them, will be working. You have to engineer that or find other ways to make it work, and that can be a significant effort.”
It’s effort the Luddy team embraces.
“In a simulation, everything is perfect,” Ali says. “When you take it to the real car, nothing is perfect. That’s the biggest challenge.”
For now, the Luddy team works with a small robot-like car that goes about 2.5 mph.
“We are doing experiments,” Ali says. “It’s a slow speed, but if it works on this one, we can make it work in the real race car.”
Testing will be done at Putnam Park Road Course near Indianapolis in April.
Liu says they will have two weeks to practice at Putnam Park and evaluate what works, and what doesn’t.
“In two weeks, you can’t do much work,” he says. “The timeline is very short.”
The June race is just the beginning. Ali says benefits include developing technologies that could improve autonomous driving in racing and in everyday life. Liu says he’s encouraged with progress that could lead to success in future Challenge races, and beyond.
“I know the whole team is pushing hard,” Liu says. “We will try our best and see what happens. If we are good, we might get a ranking, hopefully. But if not, I’m still very happy with our current rapid progress.”