Self-driving truck developer TuSimple is running a two-week pilot project
with the US Postal Service, moving mail between Phoenix and Dallas.

THE UNITED STATES Postal Service has a lot of ways to move the 484.8 million pieces of mail it handles every day. In rural Alaska, postal workers run hovercraft, prop planes, and the occasional parachute. They pilot boats in the Louisiana bayou and snowmobiles in Colorado, Minnesota, Montana, Utah, and Wisconsin. To reach the Havasupai Indian Reservation town of Supai at the bottom of the Grand Canyon, they go by mule train. And now, to carry the mail from Phoenix to Dallas, they’re letting robots do the work.

Starting Tuesday, self-driving trucks built by startup TuSimple will haul trailers full of mail and packages all by themselves. Well, mostly by themselves: The 18-wheelers will have a certified driver and safety engineer aboard, who will handle the driving on surface streets and take control from the robot as needed. The pilot project will last two weeks and include five round trips between the cities’ distribution hubs.

For the postal service, autonomy might help it reverse an ugly financial situation. The agency, which receives no tax dollars, has posted a loss every year for more than a decade. Its five-year strategic plan, covering 2017–2021, is full of talk about being open to innovative solutions. In February, it issued a request for information saying it was investigating how autonomous vehicles could fit into its fleet. It’s working with the University of Michigan on a self-driving truck to handle rural routes. In a statement, a spokesperson called the pilot “just one of many ways the postal service is innovating and investing in its future.”

TuSimple bills itself as a master of computer vision. Its cameras can see and identify threats about 1,000 meters away—more than half a mile and much farther than any lidar. TUSIMPLE


For TuSimple, it’s the chance to make some (undisclosed) revenue, pick up some press coverage, and to test its technology against the rigors of a real-world delivery service. The startup, which has headquarters in San Diego and Beijing, bills itself as a master of computer vision. Its cameras can see and identify threats about 1,000 meters away—more than half a mile and much farther than any lidar laser-scanning system. But this pilot is more of a logistical than a technological exam.

To build a high-resolution record of three freeways in three states, TuSimple executed what founder and CTO Xiaodi Hou calls an “involuntary upgrade of our mapping infrastructure.” (It used human-driven cars to do that work.) For the 1,000-mile, 20-hour drive, it had to increase the truck’s hard drive storage space to handle all the data the system will produce. And it toughened up parts of the system like the server that failed during a test ride WIRED took last December.

If that kind of problem does strike, TuSimple can’t take all day to get the vehicle back in action—not if it expects to keep its first customer happy and eventually bring in others. “These are the hard requirements that we will likely encounter in the future,” Hou says.

For this pilot, the human driver will handle the truck on surface streets, but Hou says TuSimple is already looking to a second phase of the pilot, where the robot does all the work. Meanwhile, it’s 400-person team is working to expand the robot’s operating envelope. The system can handle high winds, night driving (handling high beams was a pain), and rain, Hou says, while ice on the road remains “a nasty problem.” If it never succeeds, the postal service has a team of mules up for the task.

Via Wired