Scaling Dexterous Manipulation in Real World | Jeeyun Ahn (RLWRLD)
Jeeyun Ahn of RLWRLD discusses scaling industrial dexterity and humanoid robot deployment at the Humanoids Summit.
In this talk
- 0:29
Industrial work starts with dexterity
Ahn says RLWRLD works with industry on automation problems, focusing on dexterity because many tasks still need hands. The company analyzed human work at job sites, breaking actions into about 6,000 pieces and identifying 18 core tasks that combine into more complex work.
- 2:30
Build a data flywheel from real work
RLWRLD’s proposed cycle links data collection, model training, and deployment, with deployed robots generating further data. Its TaskBench taxonomy is intended to guide structured data collection from industrial tasks. Ahn says the company is collecting about 5,000 hours of industry data this year and expects more than 110,000 hours next year.
- 7:24
Fine-tune for deployment as technology matures
Ahn presents Real Test One, released in May, and says it ranks first on public benchmarks. He describes a gradual deployment path: robots in 2027 through 2029 will require substantial task-specific fine-tuning, since general-purpose home use remains out of reach for the industry.
- 9:48
Apply humanoids to staffing challenges
Ahn cites a Lotte Hotel example where a robot performs a task using human-collected data, and estimates the annual market at $260 million. For KDDI Lawson, RLWRLD plans to deploy 100 humanoids in 2028, starting with restocking, then aim for at least one robot in each store by 2030. At CJ Logistics, the company is working on fulfillment center tasks amid persistent staffing needs and high early turnover.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.