Robata AI is building the real-world data infrastructure needed to train the next generation of robotics and Physical AI models.

Robotics foundation models need enormous amounts of high-quality interaction data, but collecting and using real-world human data at scale remains one of the biggest bottlenecks in robotics. We are developing systems to capture, structure, and learn from human demonstrations — including video, wearable-device data, teleoperation data, multimodal sensor data, and interactions from real-world environments.

We are looking for a Research Scientist, Human Data, Robotics to help us understand how human-generated data can be captured, represented, and used to improve robotics foundation models.

This role sits at the intersection of robotics, multimodal learning, data collection, imitation learning, and human demonstration data.

We are a very early-stage startup, so we care much more about technical ability, curiosity, research taste, and willingness to build than degrees, titles, or years of experience.

What you'll work on

You will research and develop methods for turning real-world human activity into useful training signals for robotic agents.

Your work may include:

• Developing methods for learning robot behaviors from human demonstrations.
• Researching how human video and action data can be incorporated into Vision-Language-Action models, world models, and other robotics foundation models.
• Designing and evaluating different human-data capture methodologies.
• Experimenting with data collected using cameras, wearable devices, UMI-style systems, gloves, teleoperation systems, and other capture hardware.
• Building pipelines for processing, aligning, labeling, and evaluating multimodal human-action datasets.
• Studying which types of human data, sensors, viewpoints, and representations provide the most useful learning signals for robots.
• Developing methods for transferring knowledge from human actions to different robot embodiments.
• Training and evaluating multimodal models on real-world robotics datasets.
• Running experiments across simulation and real robotic systems.
• Helping define the data strategy and research direction of Robata AI.

Depending on your interests and experience, your work could touch imitation learning, reinforcement learning, multimodal learning, Vision-Language-Action models, transformers, video models, world models, robot manipulation, dexterous control, humanoids, and embodiment transfer.

Minimum qualifications