One Robot
Founding Machine Learning - World Models
$150K–$275K
San Francisco · On-site · Full-time · Posted Aug 5, 2026
About the job
Develop generative world models that simulate robotic manipulation scenes and support validating or eventually training manipulation policies without physical robots.
What you'll do
- Train action-conditioned video and dynamics models for manipulation policies, and improve long-horizon rollout quality.
- Operate multi-GPU training infrastructure, write custom CUDA, and debug large-scale runs.
- Design and improve a data engine that compounds learning across customers and manipulation tasks.
What they ask for
- Very strong coding in Python and PyTorch (or similar).
- Deep experience training image or video generation models end-to-end.
- Track record operating training runs at cluster scale.
- Working knowledge of multi-view geometry, scene reconstruction, and physical priors.
Pay
$150K – $275K (base salary)
Tools
Python, PyTorch, CUDA
AI focus
Develop generative world models for robotic manipulation, including action-conditioned video and dynamics models with improved physical fidelity and long-horizon coherence.
Model-written summary · the original posting is authoritative