Advanced self-driving lessons for future off-road robots

Researchers from Carnegie Mellon University (CMU) School of Computer Science, in Pennsylvania, have created what is thought to be the largest ever dataset to inform off-road self-driving vehicles.

In stark contrast to safety-first city tests, the high-octane study involved ragging a Yamaha Viking All-Terrain Vehicle (ATV) around a muddy, hilly test site near Pittsburgh.

A human driver guided the vehicle using a joystick and traditional braking, with the bumpy ride tracked by proprioceptive and exteroceptive sensors plus video.

The result was a dataset called TartanDrive consisting of 200,000 real-world off-road interactions covering individual wheel speeds and suspension shocks.

Self-driving scientists

Wenshan Wang, project scientist at the CMU Robotics Institute, commented: “Unlike autonomous street driving, off-road driving is more challenging because you have to understand the dynamics of the terrain in order to drive safely and to drive faster.”

Samuel Triest, a Master’s student in robotics and lead author of the team’s paper, added: “The dynamics of these systems tend to get more challenging as you add more speed.

“You drive faster, you bounce off more stuff. A lot of the data we were interested in gathering was this more aggressive driving, more challenging slopes and thicker vegetation because that’s where some of the simpler rules start breaking down.”

Over the years, many self-driving experts have predicted that “off-road applications might come first” – the logic being that private geofenced areas are more predictable environments.

This study is different, teaching AI to drive by pushing the boundaries of performance and safety in more extreme conditions.