Robotics Scientist (Policy Learning)

**About Preload** Preload builds the data layer for physical AI. Robots learn mostly from video, but human interaction is much richer: touch, force, and muscle activity. We build a wearable multimodal stack (egocentric video, tactile gloves, EMG) to capture how people actually interact with the physical world, and turn it into datasets robotics teams can train on. We're a YC (F26) team of PhD founders from Johns Hopkins and Caltech with backgrounds at Neuralink, Paradromics, and other neurotech companies. **The role** You'll prove our data makes robots better.

You'll train and deploy manipulation policies on our sensorimotor datasets and benchmark the gains against video-only baselines. **What you'll do** - Train manipulation policies (imitation learning, VLAs, RL) using our multimodal human data - Deploy and evaluate policies on real robot hardware - Design benchmarks that measure how tactile and EMG data improve performance - Feed results back to shape what we collect and how we process it - Help customers and partners get value from our datasets **What we're looking for** - 2+ years training and deploying robot learning policies (industry or research) - Strong Python and PyTorch; experience with ROS or similar robotics stacks - Hands-on experience with real robot manipulation, not only simulation - Experience with tactile sensing, dexterous hands, or learning from human demonstrations is a plus - PhD or MS in robotics, ML, or related field is a plus **Why join** Early, in-person team in San Francisco.

Your benchmarks will be the core proof point of the company.