Research Scientist

The role

You will own the end-to-end loop of Embodied AI. You will leverage multimodal datasets to develop models that generalize and self-improve in unstructured human environments.

What you will do

• Design and implement state-of-the-art robot learning algorithms for complex, high-DOF manipulation tasks.
• Architect and scale multi-modal foundation models and leverage generative video architectures.
• Own the full research-to-production pipeline from task definition to real-time deployment on physical hardware.
• Evaluate and iterate on models using the robot fleet to ensure commercial-grade performance.
• Work at the intersection of AI, Hardware, and Teleop teams to enhance Intelligence-per-Watt.

Required qualifications

• Master’s in CS/Robotics with a significant publication record in top-tier conferences (RSS, CoRL, ICRA, NeurIPS, CVPR, or ICLR).
• 5+ years of deep experience in AI and robotics research, specifically in deploying deep learning models onto physical robots.
• Hands-on experience with Vision-Language-Action (VLA) models, Video Generation, or large-scale transformer architectures.
• Mastery of Python and PyTorch (or JAX).
• Familiarity with simulation stacks like MuJoCo, Isaac, or Drake.
• The ability to write maintainable, production-ready research code that scales across GPU clusters.

Nice to have

• PhD (preferred)
• Experience with founding-stage startups or leading small, elite research teams.
• Deep understanding of classical robotics (controls, state estimation).
• Experience profiling and optimizing foundation models for low-latency inference on edge devices.

About the company

Dyna Robotics builds general-purpose robots powered by a proprietary embodied AI foundation model. The team comes from Google DeepMind, Meta, and Cruise.