Show HN: 27B open model to grade robot data 48x cheaper/27x faster vs. 6 Astra

Category: ai-ml

Tags: robotics, video-understanding, model, data-labeling

Score: 5.3/10 (Innovation: 6, Technical: 5, Documentation: 2, Utility: 5)

RoboJev appears to be a 27B parameter open-weight model for automatically grading robot task success from videos, replacing manual human labeling. The claim of 48x cheaper and 28x faster labeling is interesting for robot learning pipelines, but the provided content is only a tweet with no technical details, weights, or code.

Target audience: robotics researchers and ML engineers

Repository: https://twitter.com/josefchen/status/2113687649131336039

View on Hacker News