Job Description

Pantheon's mission is to build and deploy billions of useful robots to automate physical labor, starting with bi-manual manipulators for industry. $23M raised, 20 people from xAI, Mercor, Ramp, Midjourney, Apple, Jane Street and Zoox. This is a research-first role, not an engineering role: you'll own Pantheon's offline RL research end to end, working alongside an existing researcher on defined problems with room for your own directions.

Who can apply: Open to visa sponsorship (new H1B, TN) and visa transfers (OPT, H1B transfer). On-site in San Francisco 6 days a week.

What you'll do

• Own offline RL research end to end, from picking the question to sharing results.
• Improve policy performance on Pantheon's internal benchmarks and build on existing policy research.
• Learn from large, mixed-quality datasets (robot data, demonstrations, action-free video) under sparse rewards using goal-conditioned, contrastive and TD value learning.
• Explore model-based RL and world models for continuous control and how RL stays stable as data, compute and model size scale.
• Run clean, falsifiable experiments with reproducible baselines and see your work deployed on real bi-manual robots.

Requirements

• Published work showing strong foundational RL thinking: offline RL, value learning, policy representation (robotics, continuous control, or video/image generation backgrounds welcome).
• Research depth and taste matter far more than general MLE or data-infra skills.
• Stack: Reinforcement Learning, Offline RL, World Models, Dreamer, Isaac, MuJoCo, Brax, JAX, distributed training.
• Open to visa sponsorship (H1B, TN) and transfers (OPT, H1B).
• On-site in San Francisco, 6 days in office.

Compensation: $200K – $350K base plus competitive equity. Full-time, on-site in San Francisco. Interview: screening, technical chat with the RL research lead, 6-hour technical interview, 1–2 day work trial.

Job Tags

Similar Jobs