San Francisco’s Dimensional is hiring robot-dog “walkers” at $20 an hour to collect data
According to Kashmir Hill of The New York Times, San Francisco robotics software startup Dimensional posted a recruiting ad in a Facebook group this month seeking people to operate robot dogs walking on the street at $20 an hour.
The ad required walking 4 to 6 hours a day, with no robotics experience required, and the company provides equipment and training. The robot dogs are Unitree products and weigh about 33 pounds (about 15 kilograms). If they overheat and shut down along the way, operators must carry them back to the office. The ad has been deleted.
Dimensional is not a hardware company. It has purchased more than 100 robot dogs and 50 humanoid robots, and is developing an operating system called dimOS that can integrate with AI agents. During the walks, cameras and sensors record real-world environmental data, which is used to compare the performance of human operation and AI agent control. Founder Stash Pomichter said experiments show that agents are still not good at directly controlling hardware, and humans still need to be involved.
Original source:
Title: Wanted: Robot-dog walkers. $20 an hour Publisher: Kashmir Hill / The New York Times Publication date: 2026-10-27 URL: https://www.nytimes.com/2026/10/27/technology/robot-dog-walkers.html Source type: news report Verification status: key passages read for this release Support content: recruiting ad details, $20 hourly wage, 4-6 hours, 33-pound weight, carry-back requirement, company background and data collection purpose
Title: Kashmir Hill X post Publisher: Kashmir Hill Publication date: 2026-10-27 URL: related X post Source type: reporter’s original post Verification status: confirmed in this search Support content: core facts consistent with the NYT report
ABAB AI Insight
Dimensional is collecting “ground truth” data through human operation to improve its autonomous navigation system. The company’s comparison found that AI agents currently perform poorly at directly controlling hardware.
This approach reflects the current limitations of embodied AI in real urban environments: even when hardware can be remotely controlled, reliable autonomous navigation still requires a large amount of human demonstration data. Operators also take on walking and emergency carry-back tasks.
Similar data collection needs are common in robot software development, but appearing in the form of a public recruitment for “dog walking” has drawn attention. The company will later need to address privacy, safety and regulatory issues in order to advance the practical application of dimOS.
ABAB News · Cognitive Law
Training for autonomous navigation still relies on human collection of real-world environmental data.
Limitations such as hardware overheating mean operators must take on additional physical tasks.
AI agents still require human intervention for direct control of physical hardware at present.