Research Scientist Robot Learning VLA and WAM

Skills

About the Role

You will own vision-language-action and world-action model training from data through robot deployment. You will design action representations, adapt vision-language-model backbones for control, curate robot datasets, build action-conditioned world-model components, and use simulation, fine-tuning, and reinforcement learning to improve policy robustness.

Requirements

• PhD in robotics, machine learning, or computer vision with a robot-learning focus
• Publications, open-source work, or deployed systems
• Deep experience training modern robot-policy designs including VLA, WAM, and diffusion end to end
• Strong imitation-learning fundamentals and familiarity with RL fine-tuning
• Fluency with VLM backbones and control adaptation
• Expert Python and PyTorch skills
• Multi-node distributed-training experience with FSDP or equivalent

Responsibilities

• Own end-to-end VLA and WAM training from data to a robot-running policy
• Contribute to the technical direction for embodied research
• Close the sim-to-real gap through domain randomization, system identification, calibration, and transfer evaluation
• Adapt VLM backbones for control through encoder selection, adapters, and co-training
• Curate and weight heterogeneous robot training datasets
• Design action representation and decoding using tokenization, chunking, diffusion, and flow matching
• Build action-conditioned world-model components that predict future observations
• Run supervised fine-tuning and reinforcement-learning post-training for robustness