JOB-1811

A robotics AI startup is hiring a researcher to take charge of its offline reinforcement learning agenda, aiming to make learned robot controllers dependable outside the lab. This position is research-first rather than engineering: depth of thought and research judgment outweigh general ML engineering or data infrastructure skill. The hire joins one existing researcher, picks up problems already framed, and gets freedom to chase independent ideas.

Scope

  • Lead offline RL research from selecting the question all the way to publishing results
  • Boost policy results on the team’s private benchmarks and extend ongoing policy work
  • Learn from big, uneven-quality datasets, such as robot logs, demonstrations and video without actions, when rewards are sparse, through value learning that is goal-conditioned, contrastive or temporal-difference based
  • Explore world models and model-based RL for continuous control, along with what keeps RL stable as data volume, compute budget and network size grow
  • Design tidy experiments that can be falsified, with baselines others can reproduce, and watch the results run on real two-armed robots

Ideal background

  • PhD or MS in machine learning, computer science or a related area, concentrated on reinforcement learning
  • Papers at top ML or robotics venues such as ICML, ICLR, NeurIPS, CoRL or RSS
  • Foundational RL strength in value learning, goal-conditioned RL, policy representation and scaling stability
  • Familiarity with model-based RL or multimodal value learning
  • A path that may start in robotics and control, or in generative image or video research
  • Bonus: simulation at scale, sim-to-real transfer or open-source RL contributions

On-site in the San Francisco Bay Area. Visa transfers and new sponsorships considered.