Torc Robotics takes autonomous trucks into Michigan’s snow and ice

Those scenarios include the environmental conditions most autonomous trucking companies have deliberately avoided. Anderson pointed to “all those lovely environmental pieces that come with snow and ice and rain which otherwise you just don’t get as much as you go south around the Dallas-Fort Worth area.”

The harsh conditions feed directly into Torc’s data-driven development strategy. The company’s trucks carry an array of sensors — cameras, lidar, radar and ultrasonic sensors — designed to handle changing road conditions. That sensor data powers what Torc calls AV 3.0, an end-to-end machine learning approach to perception, prediction and planning.

“Validating our hardware and software together on public roads is a critical step in the marathon toward autonomous trucking commercialization,” said Felix Heide, head of artificial intelligence at Torc. “Each new hardware generation allows us to further validate our AI inference models, strengthen our simulation accuracy, and ensure our autonomous system performs safely and reliably in real-world conditions.”

The move required substantial state cooperation. Torc secured support from the Michigan Economic Development Corporation (MEDC), Michigan Department of Transportation (MDOT) and Ann Arbor SPARK.

“Torc’s continued growth in Michigan highlights the importance of strong public-private partnerships in advancing next-generation mobility,” said Quentin L. Messer Jr., CEO of the MEDC. “Through collaboration with companies like Torc, Michigan is driving innovation, building a skilled workforce, and reinforcing its position as a global leader in autonomous and connected vehicle technologies.”

Anderson emphasized that state-level backing proved essential. “When you look at the broad support from the Michigan Economic Development Corporation financially and the added ability of cooperation from Michigan Department of Transportation for operation, it just provides a greater ability for Torc to evolve both its hardware and software with the support of the state infrastructure like that.”

Central to the effort is Torc’s transition to AV 3.0, an end-to-end machine learning approach that represents a fundamental shift from earlier development. Anderson described the previous AV 2.0 as “essentially just a black box, a monolithic black box.”

“The real crux of AV 3.0 is moving away from that to an approach which basically covers perception, planning and prediction through components that essentially are built with end-to-end models that otherwise when we bring them all together can be trained with one neural net,” Anderson said. “Being exposed to harder environments or harsher environments plays into the learned behavior of what we’re doing.”

The modular architecture allows individual components to be decomposed, validated and updated independently — critical for a data-driven system that improves over time.

Anderson added, “Just like a lot of other things that are machine-learning-based, they get better over a period of time.”

Torc is actively hiring in Michigan and remotely for software engineering, artificial intelligence and machine learning roles. The company targets commercialization in 2027, with Anderson framing the Michigan expansion as part of “our maturing path toward taking a driver out of the vehicle.”