DARPA returns to off-road environments in latest autonomous vehicle project

The inhospitable terrain of the Mojave Desert provided a formidable challenge for embryonic autonomous-vehicle technology during the DARPA challenges of the mid-2000s.

Now the federal agency behind those formative competitions has plans for further testing AV tech in similar off-road environments, with the latest developments coming in simulation.

The Defense Advanced Research Project Agency, better known by its DARPA acronym, on Wednesday granted Intel and its collaborators a four-year contract to conduct research and simulate unusual topography.

“Off-road environments present different types of challenges, including uneven terrain, cliffs, mud, large rocks, bushes,” says Germán Ros, director of the Autonomous Agents Lab at Intel Labs. “In this project, we’re creating a simulation platform that can accurately generate all types of extreme situations, helping autonomous ground vehicles learn how to effectively navigate off-road environments at high speeds.”

This aspect of the DARPA work, part of the “Robotic Autonomy in Complex Environments with Resiliency,” or “RACER” program, will combine researchers from Intel Labs along with researchers at the Computer Vision Center in Barcelona and the University of Texas.

Intel says there’s no connection to the on-road autonomy work being done by the company and its Mobileye subsidiary, which are focused on near-term deployments of robotaxis. But broadly, lessons learned across on-road testing will help guide the off-road project for uses cases farther in the future.

“This project seeks new ways to transfer the key ideas used in on-road environments, such as simulating edge cases, situations that could be challenging for an autonomous system, all adapted to off road,” said Ros, who previously worked at the Toyota Research Institute.

No teams completed the 132-mile desert course between Barstow, Calif., and Primm, Nev., during the DARPA Grand Challenge in 2004. A year later, five teams completed the feat on a similar route.

In simulation, Intel Labs says “massive” new set of environments will allow researchers to cover more than 100,000 square miles, accelerate the gap between on-road and off-road vehicle deployments and reduce costs typically associated with testing in real-world environments.

Whereas the first set of DARPA autonomous-vehicle challenges helped hone basic technology, the ongoing RACER program is designed to enhance the ability of AVs to perform at the edge of their performance envelopes and, in some cases, beyond what’s capable at human-driven speeds.

Real-world testing is also part of the program. Last October, DARPA selected Carnegie Mellon University, NASA’s Jet Propulsion Laboratory and the University of Washington to participate in collecting data from and integrating autonomous tech into high-performance all-terrain vehicles.