Founding Research Engineer, Robot Learning
Research · Full-time · San Francisco · On-site
You'll build and train the models that learn from physical experience — the vision-language-action policies and learning algorithms at the core of what we do — and deploy and validate them on real robots, not just in simulation. This is a research-and-engineering hybrid: you'll move fluidly between inventing methods and shipping the code and infrastructure that makes them work.
What you'll do
- Develop and train robot-learning and vision-language-action models for manipulation and control.
- Deploy and validate policies on real robots, and feed field failures back into training.
- Own the end-to-end loop: data collection, training, evaluation, and deployment.
- Build data-collection pipelines and curate large-scale datasets and training recipes.
- Build evaluation infrastructure that ties training metrics to real-world robot performance.
- Run fast, rigorous experiments, diagnose failure modes, and iterate quickly.
What we're looking for
- Strong Python and a modern deep-learning framework (PyTorch, or JAX).
- Hands-on experience developing and deploying robot-learning systems on real robots — not simulation alone.
- Experience with imitation learning / behavior cloning and/or reinforcement learning for control.
- A background in robot manipulation or visuomotor policies.
- Comfort moving fluidly between research and production engineering.
Nice to have
- Publications at CoRL, RSS, ICRA, or NeurIPS.
- Experience with vision-language-action models and diffusion policies.
- Large-scale / distributed training (FSDP, DeepSpeed, Ray).
- Familiarity with simulators (Isaac Sim, MuJoCo) and ROS / ROS2.
Email us at eldaniz@episodeint.com.