Autonomous systems learn from modelling and simulation, including self-driving cars, uncrewed aircraft, and humanoid robots. However, the training process can take months to years and has trouble accounting for all the uncertainty in the real world. In robotics, this is known as the simulation-to-real gap.
The Defense Advanced Research Projects Agency (DARPA) has implemented the Transfer Learning from Imprecise and Abstract Models to Autonomous Technologies (TIAMAT) program to improve this gap. DARPA has selected Indy Autonomous Challenge as an official test and evaluation platform for the TIAMAT program, which aims to develop rapid autonomy transfer techniques to enable same-day autonomy that is robust to the quick and inevitable changes in dynamic environments and adaptable to a variety of platforms and domains.
We were inspired to create the Indy Autonomous Challenge by the successes of the DARPA Grand Challenges, held 20 years ago and gave rise to the modern autonomous vehicle industry,” said Paul Mitchell, CEO of Indy Autonomous Challenge. “It is an honour to partner with DARPA to accelerate the development and training of Physical AI using our first-of-its-kind robotics platform of the world’s fastest autonomous racecars.”
There is no question that AI will revolutionize autonomy in the physical world, but testing this in the real world is expensive and risky. The IAC SIM to REAL platform will allow rapid iterative testing of novel AI models and piece together those that work best in high-speed and edge-case environments.
Over the next three years, IAC will work with DARPA to build a SIM-to-SIM and SIM-to-REAL test and evaluation platform that will dramatically improve the speed and efficiency of AI driver training.
Other TIAMAT performers, as well as current IAC University Teams, will be able to test various AI models in low—and high-fidelity simulation. Those proven to work will graduate to real testing on a race track in IAC AV-24 racecars. This AI training platform will not only benefit high-speed ground vehicles but could also be applied across other autonomous domains such as air, sea, and space.