Uber’s self-driving cars are heading south for the winter, in a manner of speaking. The ride-hailing giant today revealed that its Advanced Technologies Group will kick off manual driving in Dallas, Texas this November, with the goal of developing high-definition maps and capturing scenarios that can be recreated in simulation and on test tracks. It’s a precursor to driverless car deployment on public Dallas roads, but Uber says it may choose not to test its self-driving systems right away.

In the near term, teams of specially trained drivers will rove around downtown Dallas, helping to verify that Uber’s autonomy system would perform as intended were it deployed. Uber says it’ll partner with “key organizations and stakeholders” to both “educate” and generate awareness for its efforts in Dallas, and it says that if it decides to operate vehicles autonomously after this data collection phase, it’ll work with stakeholders at both the state and local level.

“Self-driving technology has the potential to create even more value to our users [and] we are … committed to ensuring that every mile we drive on public roads contributes meaningfully to our development work,” wrote Uber in a Medium post. “In order to arrive at that future, we must approach building this technology thoughtfully and with a strong sense of responsibility to the communities where we operate, which our team is dedicated to doing every day.”

In an S-1 filing ahead of its initial public offering earlier this year, Uber noted that its Advanced Technologies Group has grown from a team of 40 Pittsburgh-based researchers in 2015 to a 1,000-person workforce spread across offices in San Francisco and elsewhere. Furthermore, it said that it’s collected data from “millions” of autonomous vehicle testing miles to date and completed “tens of thousands” of passenger trips to date.

Looking ahead, Uber expects that cars like those it’s currently testing will replace drivers in some situations, but not right away. Instead, it’s predicting a “long period of hybrid autonomy” in which autonomous vehicles are deployed “gradually” against specific use cases — for instance, in scenarios involving substantial traffic, complex routes, or unusual weather or during “high-demand events” such as concerts or sporting events.

Uber isn’t naive — it concedes that its investments in autonomous vehicles may “add to driver dissatisfaction over time” — but it says that it expects that drivers will “[remain a] critical and differentiating [market] advantage” for it and continue to be “valued partners” for the long term. That said, Uber also predicts that autonomous vehicles will “substantially” reduce the cost of providing ride-sharing, meal delivery, or logistics services.

Uber has a mixed track record when it comes to self-driving car research, to put it mildly. It restarted tests of its driverless cars in Pittsburgh last December, eight months after one of its prototype Volvo SUVs struck and killed a pedestrian in Tempe, Arizona, and began manual tests in San Francisco and Toronto.

This summer, the National Transportation Safety Board (NTSB) determined that Uber engineers had disabled the automatic emergency braking system in the Volvo XC90 involved in the accident. (In internal documents, the company said this was to “reduce the potential for erratic vehicle behavior.”) The NTSB also found that the car’s perception system detected the victim about six seconds before impact but that it didn’t determine emergency braking was needed until 1.3 seconds before impact.

Following the crash, Uber halted self-driving tests across the country and laid off 100 of its autonomous vehicle operators.

In a voluntary safety assessment filed with the NTSB, Uber said that with a newly established systems engineering testing team it’s now better positioned “to reason over many possible outcomes to ultimately come to a safe response,” and that it intends to form a self-driving safety advisory board of outside experts. It also said that it spent months testing its self-driving technology on a closed track and completed a lengthy internal review, and that it hired an adviser — former NTSB chair Christopher Hart — to assess its safety culture.

In a blog post published in June, head of ATG Eric Meyhofer detailed newly implemented safeguards, such as a training program focused on safe manual driving and monitoring systems that alert remote monitors if drivers take their eyes off the road. It now mandates that teams of two employees — “mission specialists” — drive its cars, switching off every two hours. And in Pittsburgh, Uber says that those cars operate on a mile-long route between two of its offices in the city’s Strip District and travel no faster than 25 miles per hour, and stay off of the roads at night and during inclement weather.

The bumps in the road haven’t deterred investors. Earlier this year, Reuters reported that group led by SoftBank and Toyota is in talks to invest $1 billion or more into Uber’s self-driving vehicle unit, which would value the unit at $5 billion to $10 billion. It might have competition in spades — Waymo, Cruise Automation, Tesla, Apple, Zoox, Aptiv, May Mobility, Pronto.ai, Aurora, and Nuro, to name a few — but in a global market that’s forecast to hit revenue of $173.15 billion by 2023, there’s plenty of cash to go around.