What is Autonomous Racing?
No driver, just software. Driverless vehicles are the vehicles of tomorrow. Autonomous racing is the ultimate engineering challenge. The driver is replaced with a variety of sensors that act as the eyes and ears of the vehicle, feeding the data to a control algorithm that students develop and tune. Machine Learning (Artificial Intelligence) is often used to interpret sensor data.
The official vehicle of the Indy Autonomous Challenge is the Dallara-built AV-24, the World's Fastest Autonomous Racecar, introduced at CES 2024 to replace the original AV-21. The AV-24's Dallara chassis is a modified version of the classic "Indy Lights" chassis design.
Driverless: Pure Software
Autonomous racing replaces the driver with a variety of sensors that act as the eyes and ears of the vehicle. The sensors take in vehicle and environment information, process the data, and send command information to the control algorithm. Within the control algorithm the autonomous racecar senses where it is on the road, what is around it, and what path it should follow next.
The software has three parts: first is control, the low-level hardware managing steering, braking, gear-shifting, and acceleration. Second is localization, using GP, IMU fusion, Kalman filtering, and pose estimation. Third is perception, using LIDAR and other sensors to ingest, fuse, and filter, and also for world state estimation.
Ultimate Engineering Challenge
The challenge of developing a vehicle that races autonomously adds a multitude of requirements for students to take into account when tuning a control algorithm. The most important challenge is the fact that, unlike a normal vehicle whose speed rarely exceeds 80 mph on a public road, a racecar can go upwards of 200 mph.
This is all accomplished with 3 Luminar Hydra LiDAR pushing out 64 Lines and a 60-degree scanning window, 2 Novatel GPS with millimeter accuracy, 6 high definition wide-field cameras, 3 RADARs located on both sides and the front of the car, and a state-of-the -art GPU and processing computer.
Because of these high speeds and mass amounts of data, students must carefully interpret sensor data to create the most efficient control algorithm, as a racecar at top speed of 200 mph can travel about 293 feet (90 meters) a second!
Vehicles of Tomorrow
Self-driving vehicles will revolutionize the way we all travel and perceive transportation infrastructure. Autonomous Technology merges robotics, machine learning, engineering, and modern software development methods.
When all of this comes together, it pushes the boundaries of consumer technology. Developing a racecar that can stop or avoid a collision, which going 200 mph, makes road vehicle algorithms that much more robust - adding a huge margin of safety to consumer autonomous vehicles that would otherwise never be developed.
Under the Hood: Students & Team Leads Share Insights
“The approach we’re using is best described as neuro-symbolic AI, a hybrid version becoming more common for increasing the robustness of AI in many industries.... The idea is, you feed neural inputs into a mostly rule- and structure-based algorithm, to enhance its robustness. The main advantage of this approach is that it’s no longer a huge black box, where we don’t know why something different than what we expected came out. Since the core decision-making aspect of our data pipeline is rule-based, we can see the input and know what’s coming out, because we coded it.”
- Sid Saha, Senior Technical team Lead & Planning Team Lead
"The essence of controls lies in the ability to adapt swiftly to ever-changing system dynamics. Our cutting-edge research focuses on optimizing runtime and speed, ensuring that our autonomous systems not only navigate complex environments, but do so with precision and efficiency. Our team uses advanced learning methods that enable us to quickly adapt to evolving system dynamics, ensuring mathematical convergence and real-time exploration of past trajectories. This approach not only guarantees safety but also fuels continuous learning from the environment."
- Haoru Xue, Senior Technical team Lead & Controls Team Lead
“To ensure the utmost safety in the future of autonomous vehicles, it is absolutely crucial that we deliberately push to the absolute limit... Pioneering research involves finding and understanding weaknesses in the system, and constantly improving and advancing engineering techniques... In this competition, we are operating on a unique cutting edge of autonomous research, unlike anything else in the world right now.
"While developing autonomous systems, we meticulously balance real-world testing with extensive simulation training, data generation, and iterative testing. For every hour we spend in real-world scenarios, we dedicate at least 100 hours to simulations, ensuring that our technology is thoroughly tested, refined, and ready to meet the challenges of the road ahead."
- C.K. Wolfe, Simulation Team Lead