At 1:16 on a Wednesday morning, a truck left northeast Houston carrying a paying customer's freight. It arrived near Dallas at 4:57. For most people, this would qualify as an unusually dull night. For Xiaodi Hou, the interesting detail was the seating arrangement: nobody was in the cab.
The April 29, 2026 journey covered roughly 230 miles to Hutchins, Texas. Bot Auto, the company Hou founded after leaving TuSimple, said the truck completed the delivery without a safety driver, an onboard observer or low-latency remote steering. Ryan Transportation had booked the load through the ordinary machinery of freight brokerage. An autonomous vehicle had accepted an ordinary job.
Hou has spent years trying to make those two things meet. His first trucking company produced a public listing, a driverless milestone and a spectacularly unhappy separation. His second has given him another chance to answer a question that a moving truck cannot settle by itself: can the business carry its own weight?
The trouble with a beautiful demonstration
Hou co-founded TuSimple with Mo Chen in 2015. As chief technology officer, he helped turn computer vision research into systems intended to guide heavy trucks along public roads. The ambition required hardware, software and the inconvenient participation of a world that would not sit still for the cameras.
In April 2021, TuSimple went public. That December, its truck completed an approximately 80-mile driver-out journey from Tucson toward the Phoenix area. The vehicle negotiated surface streets and highways without a human aboard. For an engineer whose work had once concerned what a computer could notice in an image, this was a rather large object to set in motion.
The test truck had a name: Ghost Rider. The comic-book flourish seems almost restrained beside the corporate events that followed. Hou became TuSimple's president, CEO and chairman in March 2022. By the end of October, the board had removed him.
Looking back in 2024, he described the refuge he had made for himself inside the technology function as a “very comfortable cocoon.” It is a revealing admission from a founder. Technical authority had allowed him to concentrate on the problem he liked best, while the company surrounding it developed problems of its own.
“very comfortable cocoon”
Hou's description of his earlier life as CTO
The phrase gives his second company a useful backstory. He had already experienced the satisfaction of making a truck drive itself. He now wanted to attend to the parts of the enterprise that do not appear in a driving video: the cost of acquiring customers, maintaining equipment and keeping a working fleet busy.
Before the highway, the problem of seeing
Hou's route into trucking began with a more fundamental puzzle: how do people find the interesting object in a complicated scene? He earned his engineering bachelor's degree at Shanghai Jiao Tong University in 2008. A year earlier, he and Liqing Zhang had published a paper on detecting visual saliency using a spectral residual approach.
The terminology is formidable; the everyday experience is familiar. Look at a photograph and something catches your eye before you have consciously inspected every object. Hou's research explored computational ways to identify such regions. The work concerned how visual information could be reduced to something useful without first naming everything in sight.
At Caltech, he pursued computation and neural systems under Christof Koch. His 2014 doctoral thesis examined vision, including saliency, object segmentation and the difficulties of evaluating boundary detection. It questioned some established benchmarks as well as proposing algorithms. Even in the laboratory, deciding what counted as a good result was part of the problem.
After his doctorate, Hou founded CogTu, which applied image recognition to advertising. TuSimple followed. The commercial setting changed, but the scientific thread remained visible: turn the contents of an image into information a machine can use. A truck added motion, physical consequences and customers with delivery appointments.
There is a smaller, pleasantly incongruous detail in his public biography. His LinkedIn education entry lists theater arts and marathon among his activities. It also includes a self-reported Erdős-Bacon number of six, combining chains of mathematical collaboration and acting connections. Very few trucking executives offer an academic route to Paul Erdős alongside a theatrical route to Kevin Bacon.
A departure that refused to stay in the past
Hou's exit from TuSimple was contested. The board's removal followed an internal investigation into dealings involving Hydron, a separate truck venture associated with Chen. TuSimple said it had lost confidence in Hou's judgment. Hou disputed the grounds for his removal. He retained a board seat for several months before resigning in March 2023.
The dispute later extended to Bot Auto. TuSimple alleged that the new business had misappropriated trade secrets involving sensing, decision-making and safety technology. Hou and Bot Auto contested the accusations. An engineer could begin a new codebase; a founder could not so easily leave the old company behind.
On September 15, 2026, a Texas appeals court affirmed the refusal to grant CreateAI, TuSimple's renamed successor, a temporary injunction against Bot Auto. That was a ruling on preliminary relief, rather than a final determination of the underlying claims. Its immediate significance was practical: the requested restriction on Bot Auto would remain denied.
The legal history complicates any tidy comeback narrative. Hou's new venture shares a field, a founder and some professional connections with the old one. His task has included explaining why a second attempt deserves to be assessed on its own work. The courtroom and the loading dock impose rather different standards of proof.
Houston supplies a less glamorous question
Founded in 2023 and publicly unveiled in September 2024, Bot Auto began its public life with $20 million in pre-Series A funding. Houston was its base. The company would own and operate its trucks, selling transportation capacity to freight customers. Hou was taking responsibility for what happened after the autonomy system was installed.
This choice changes where the awkward questions land. If a truck is unavailable, a customer still needs a shipment moved. If a new capability requires expensive equipment, somebody must pay for it. A carrier has to reconcile the engineering decision with the bill, the schedule and the next load.
By July 2025, Hou was describing cost per mile as his central commercial measure. His argument was straightforward: excellent autonomous performance would have little commercial value if moving the freight still cost more than the available alternative. Maintenance, operating efficiency and the technology itself all belonged in the calculation.
There is a certain comedy in spending years teaching a machine to negotiate traffic, then discovering that the next examination resembles bookkeeping. But Hou's emphasis makes the ambition legible. A customer can evaluate a price and a delivery. It need not become an expert in neural networks to buy transportation.
His public discussions also keep human work in the picture. He has described a future in which autonomous trucks handle highway freight while people take on other parts of transportation, including local deliveries and customer interaction. Automation may change the division of labor; the shipment still passes through an organization full of people.
An empty cab, with conditions attached
The overnight trip to Hutchins gave that argument a concrete setting. The truck left a parking facility in northeast Houston, carried a standard commercial booking and arrived south of Dallas before dawn. The company's definition of humanless excluded an onboard observer and a remote driver providing low-latency control.
Those details matter because an empty driver's seat is only one way of describing an autonomous operation. A passenger watching the road or a person steering remotely would introduce different operating requirements. Hou has pushed for clearer definitions so that customers and competitors can judge the substance behind a milestone.
No human in the cab or low-latency remote steering, according to Bot Auto.
One completed load establishes a specific accomplishment. Repeated service asks further questions: how consistently can the system operate, how often does it need assistance and how much does the whole arrangement cost? These are the questions Hou's commercial emphasis invites. Success makes the examination longer.
It also makes his story more interesting than a simple race to remove a driver. He has reached a driver-out milestone before. This time, the distinguishing promise is that the equipment, operating model and customer demand will fit together well enough to sustain a company.
The next bill arrives with the hardware
In October 2026, Bot Auto selected Lenovo's AD1 controller, built on NVIDIA Hyperion, for its production computing platform. Integration was scheduled to begin in the first quarter of 2027. The company said it expected the autonomy hardware, including computing and sensors, to cost $50,000 per truck.
That is a company estimate for the autonomy system, rather than the purchase price of a complete truck. Bot Auto also reported approximately 350 loads a month across Texas. Its announcement did not identify that entire volume as humanless. Such distinctions are necessary when a business is moving between development and a larger commercial fleet.
Hou's more recent writing looks beyond the vehicle to the institutions around it. In August 2026, he considered aviation as a precedent for physical AI, asking how safety, economics, rules and financing develop together. His attention had expanded from the machine to the market that must accommodate it.
That expansion is the revealing part of his second attempt. A researcher trained to distinguish objects from their surroundings has had to reckon with how much the surroundings matter. The truck needs roads, customers, reliable equipment and a company capable of organizing them.
At the end of the overnight journey, there was freight to hand over. For Hou, that ordinary obligation is now a demanding ambition. The next load will need the same attention. It has very little interest in a founder's reputation.