@ComoxAIx

We provide high-fidelity real-world datasets for embodied AI and robotics, closing the sim-to-real gap with deployment-specific data collections.

 Vancouver, Canada
comoxai.com
Joined April 2026
  • Tweets5
  • Following40
  • Followers2
1Media
The data your models actually need. Synthetic data breaks in reality. We capture high-fidelity, deployment-specific real-world datasets for embodied AI & robotics — closing the sim-to-real gap. Built for deployment. comoxai.com #EmbodiedAI #Robotics #SimToReal
1
46
Hard eval data for VLMs and multimodal models Frontier VLMs are often good enough on clean benches and still fail on messy real footage, exact object counts, rare conditions, long-tail scenes public data and sim cover poorly. Useful ≠ accurate under those conditions.
2
1
17
You get a scoped eval split, labels, a short protocol/failure breakdown, reach out at [email protected] For teams evaluating VLMs / multimodal / physical perception. Not a teleop farm or freelance videographer mill.
5
Comox collects real-world video that’s hard to get from the open web, annotates with multi-reviewer QC, and delivers hard eval sets. On our hardest tests, frontier models have scored under 20%, and 0% when they must count objects exactly.
12
Unscalable data makes scalable AI. Right now, our fleet is capturing high-density edge cases in dynamic physical environments to train a next-generation autonomous perception model. It is highly manual, unscalable grunt work. And it is exactly why the dataset is so valuable.
1
36
Load more