Research Scientist, Vision-Language-Action Models

Matter • Menlo Park, CA

2d • Onsite

About The Position

Matter is building the AI-native autonomy stack for physical manufacturing in the United States. We operate as a contract manufacturer, deploying software and autonomy in our own factories, which gives us something most AI companies don’t have: a live production environment as a training ground. Our long-term vision is to become the infrastructure layer for American manufacturing, the way AWS became infrastructure for software. The role is to lead the development and deployment of Vision-Language-Action (VLA) models for robotic manipulation in live manufacturing work cells. This is not a lab role. You will train models, close the Sim2Real loop, and deploy them on physical robots running production programs. Matter’s Sim2Real pipeline spans NVIDIA Isaac Sim, physics-accurate virtual builds of our modular assembly equipment, and 100% data collection from real factory operations. You will operate at the center of this flywheel design, improving models with every production run.

Requirements

• PhD, Graduate or equivalent research depth in robotics, machine learning, or a related field
• Hands-on experience training and deploying VLA, VLM, or generalist robot policies on physical hardware (not just simulation)
• Strong foundation in imitation learning, reinforcement learning, and general machine learning methods
• Proficiency in PyTorch; experience with NVIDIA Isaac Sim, MuJoCo, or similar physics engines
• Ability to debug the full stack: model architecture, training data quality, sim calibration, sensor noise, and hardware edge cases
• Comfort operating in a high-velocity, ambiguous environment where you own systems end-to-end

Nice To Haves

• Experience with MARL or multi-robot coordination
• Background in manufacturing, industrial automation, or robotic assembly

Responsibilities

• Develop and fine-tune VLA models for precision assembly tasks, including dexterous manipulation, part handling, and test operations
• Design and manage the Sim2Real training pipeline: domain randomization, synthetic data generation, physics simulation (NVIDIA Isaac Sim, MuJoCo), and sim-to-physical transfer
• Build evaluation frameworks to benchmark real-world manipulation performance against manufacturing tolerances and repeatability requirements
• Collaborate with controls and automation engineers to fuse learned policies with traditional control architectures for production safety
• Contribute to the Physical AI architecture decisions: model selection, data strategy, training infrastructure, and deployment protocols
• Publish novel research in top tier 1 conferences — though shipping production systems is the primary measure of success

What This Job Offers

• Job Type: Full-time
• Career Level: Senior
• Education Level: Ph.D. or professional degree