San Francisco startup Dimensional advertised for people to walk robot dogs through city streets, helping train systems intended to operate autonomously. The job is a useful glimpse of physical AI in progress: today’s robots can still need a human at the controls, and occasionally a lift home.
Watch Desk analysis
What happened
Digital Trends reports that the now-deleted advert sought people to walk the dogs for four to six hours, with no previous experience required. The walkers steered the robots by remote while their cameras and sensors recorded the journeys. The machines were $2,000 Unitree robot dogs, according to the article.
Dimensional is developing dimOS, an open-source platform intended to let AI agents operate different robots through natural-language instructions. The article says its public repository labels the platform a prerelease beta and documents remote operation through a browser or Meta Quest headset.
Why it matters
Those walks are not just an unusually elaborate dog-walking shift. They give Dimensional examples of human navigation that could help improve its autonomous systems. But collecting demonstrations is not the same as getting a robot to handle unfamiliar places and tasks reliably without help.
The article also describes Anthropic research in which tested language models paired with trained controllers performed worse than the controllers alone on familiar simulated object-handling tasks. That finding is specific to the tests described, but it underlines a central challenge: a capable language model does not automatically make a capable robot.
Our read
The interesting development is the bridge between language-based agents and real machines, not the novelty of a robot dog on a pavement. Dimensional’s platform ambition is clear; the walkers’ work shows how much human guidance still sits behind it. Physical AI may yet save labour, but the training phase is not above borrowing a pair of walking shoes.
What to watch
- Whether Dimensional moves dimOS beyond its prerelease beta and shows what robots can do with it.
- How much of the recorded navigation data improves autonomous performance, rather than just building a larger dataset.
- Whether future demonstrations show robots completing unfamiliar tasks with less human steering.
Discussion spark: When training robots for real-world work, is human-guided data collection a sensible bridge to autonomy, or a sign that the promised labour savings are still a long way off?
Sources and evidence
- Source update (27 October 2026, 11:17 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.