Banseog note · 2026-10-26
A company building a 'foundation model' for robot hands
Current stage · from public events · 2026-11-14
- R&D · prototype
- Pilot · first production
- Scaling production
- Operations · field rollout
- Going global · localizing
- M&A · integration
- Restructuring
R&D · prototype Published GEN-1.5 research results on learning new tasks from short demonstrations
- 2026-08-19 The company published GEN-1.5 one-shot and few-demonstration learning experiments, explicitly noting that the test tasks were limited to short and simple tasks. Source
Roles often seen at this stage Robot software, AI & vision · Controls & embedded · Mechanical & hardware design
Companies at the same stage → · Same stage in other industries →
- In June 2026 it announced a new USD 400 million investment led by Radical Ventures. 8VC, Union Square Ventures, NVIDIA and others took part, and the company said its total funding had passed USD 500 million.
- The company said GEN-1, released in July 2026, was trained on more than 500,000 hours of real interaction data and tested on about 9,000 end-effector (robot hand and tool) variants.
- The company said GEN-1.5, released in August 2026, learns a new task after watching a single 3–12 second demonstration, recording an average one-shot success rate of 59% across 10 tasks, and 83% with additional training on five minutes of data.
Decisions within a year
- In June 2026 it said it would use the USD 400 million to scale robot learning, develop next-generation models, expand its physical data collection system and build compute and training infrastructure.
- It released GEN-1, which handles many different robot hands and tools with a single model, on July 23, 2026, followed by GEN-1.5, which is capable of one-shot learning, on August 19.
- It is currently hiring for 22 roles at two sites, San Francisco and Boston, including pre-training and post-training researchers, ML infrastructure, robot control software, mechanical and electrical engineers and data partnership staff.
People they will look for
- Robot control software engineer — for one model to handle thousands of robot hands and tools, the control layer has to be refined on real hardware.
- ML infrastructure and training optimization engineer — a large share of the funding goes to compute and training infrastructure, and more than 500,000 hours of data must be trained on continuously.
- Data partnerships and robot operations lead — following the decision to expand physical data collection, more data collection sites and partners are needed.
For hiring managers
In Korea, conglomerate robotics research organizations, humanoid startups and the Korean arms of global AI companies look for the same robot learning and ML infrastructure talent. Hiring is centred on on-site work in San Francisco and Boston, so for Korean candidates the starting point is whether they can relocate.
For job seekers
Experience collecting and training on robot manipulation data at scale, or operating training infrastructure for large models, is what this company values most right now.
An analysis written from public sources. Sources: Generalist blog 2026-06-04 (USD 400 million investment, investors, use of funds), Generalist blog 2026-07-23 (GEN-1, 500,000 hours, 9,000 end effectors), Generalist blog 2026-08-19 (GEN-1.5 success rates), Generalist careers page (22 roles in San Francisco and Boston).
Roles hiring now
Skills asked for most
Open jobs
Banseog checks official company careers pages every day. A job is marked closed only when the official page returns a removal (404 or 410).