RoboBoom
Machina Summit· July 22, 2026· 16:21

Dan Carpenter (Uber AI) & Sandor Felber (Minerva Humanoids): Training the Machine | MACHINA 2026

Dan Carpenter, global head of Uber AI at Uber, discusses robotics data collection and AI services with Sandor Felber of Minerva Humanoids at MACHINA 2026 in Paris.

In this talk

  1. 1:14

    Uber’s data business

    Carpenter says Uber AI Solutions has spent three years offering services including data labeling, mapping, autonomous vehicle data, model evaluation, and reinforcement learning environments. Uber’s network of 9 million earners helps it gather data at scale, including unusual cases useful for robotics.

  2. 3:24

    Collecting useful robot data

    Felber describes data collection, hiring, and building a core system as major challenges for robotics companies. Carpenter argues that physical robotics lacks the vast public training material available for language models, and says collection must balance scale with quality across settings such as homes and industrial environments. Felber adds that teleoperation data takes substantial effort to clean and prepare.

  3. 5:26

    Workers, infrastructure, and edge cases

    Uber applies parts of its driver platform to data work, including identity checks, payments, and onboarding; Uber AI Solutions operates in 30 countries. Carpenter highlights difficult driving conditions and places without street names as examples of valuable edge cases, and says diverse locations, homes, and collection methods can help avoid relying on repetitive datasets. He also cites uncomfortable headsets, slow uploads, and quality review as operational bottlenecks.

  4. 9:08

    Where physical AI could grow

    Carpenter says Uber’s autonomous vehicle partners collect external sensor data, building large sets of observations across different traffic and lighting conditions. Uber is exploring collection through backpacks and delivery work. He points to agriculture, including weed control at his brother’s winery using Carbon Robotics, as a promising robotics application, and sees potential in both wheeled and legged robots.

Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.

More talks