This Robotrucker Says Its AI Drives Highways Without Advance Training
Conventionally, AI drivers have been giant travel nerds. The models are extensively trained on the routes they take, using extensive simulation and high-definition mapping to ensure they are thoroughly familiar with the route before they take off. But Canada’s Waabi said its AI-enabled trucks can now handle lengthy, fully automated highway runs without any such advanced training.
The Toronto-based company, which is getting ready to start commercial robotruck operations next year, said it hit an industry milestone last week by having one of its test trucks drive about 300 miles on the highway from Dallas to San Antonio without having previously tested the route in simulation, done any practice runs or been given new real-world data. Essentially, the truck drove as a human would on an unfamiliar route. The safety driver at the wheel also didn’t need to take control at any time during the test run, Waabi founder and CEO Raquel Urtasun told Forbes .
The truck “was able to drive perfectly, confidently, safely, with no interventions for the entirety, without ever seeing that lane before. With no data, no engineering, no retraining, no nothing, instantaneously,” she said. “ When you think about what that means in terms of meeting the customers’ needs, this is a massive unlock of self-driving truly to get to scale.”
The test run with a generalized AI driver took about four and a half hours last week, from early to late morning, she added. Speeds averaged around 60 miles per hour.
If Waabi can safely deploy trucks using its tech on new, unfamiliar routes, it would give the company an edge over competitors in the segment, such as Aurora, that are building up robotic truck operations slowly on familiar, well-understood routes. Even Waymo, the industry leader in autonomous driving–and which suspended its own trucking program a few years ago–spends weeks or months mapping and simulating how its robotaxis will navigate new cities before picking up passengers in order to ensure smooth rides. Elon Musk’s Tesla, which is expanding its robotaxi program in Austin and other cities, claims its AI doesn’t need high-definition mapping and extensive practice to handle new routes. However, it appears to rely heavily on remote human monitors to avoid problems.
Founded in 2021 by AI scientist Urtasun, Waabi’s goal has always been to develop a generalized AI driving system that adapts to unfamiliar situations as easily as human brains do, making reasonable and safe decisions based on available information. It’s working with truckmaker Volvo, which expects to start installing Waabi’s tech on new, purpose-built self-driving trucks in the first quarter of 2027. Backed by Uber, the company raised up to $1 billion this year from investors including Khosla Ventures, G2 Venture Partners, the Volvo Group and Porsche.
“If you look at freight today, it really moves in a very dynamic way. It constantly shifts in terms of where goods get delivered for customers,” Urtasun said. “When you think about the way that autonomous trucking should work, that autonomous truck needs to go where the customers want. It’s not that the customers need to go where autonomy wants them to be.”