Physical AI Scientist for Welding & Smart Manufacturing

Amazon Science; Boston (MA); On-site; USD 143,000 - 193,000; Full time

Job summary

Amazon Science is seeking an Applied Scientist to create first-of-its-kind physical AI for the OMHS Software Controls and Science team. You will turn applied research into prototypes and production systems that operate on real parts on shop floors, combining ML, 3D perception, and geometric reasoning to match parts to CAD models and determine tooling without manual programming.

You will own the full loop from sensor setups to model training and production integration, enabling a lossless digital

Qualifications

• PhD in engineering/CS/ML/robotics or equivalent quantitative field.
• 1+ years of deep learning, computer vision, algorithms implementation experience.
• Hands-on experience deploying robots or autonomous systems in real-world environments.
• Ability to debug the full stack from perception to control.

Responsibilities

• Deploy 3D perception and part-recognition systems on robotic welding cells matching CAD models and determining orientation, tooling, fixturing.
• Build software pipelines connecting CAD interpretation to robot motion planning and execution in a production environment.
• Frame fabrication problems as scientific problems and prototype end-to-end solutions from data collection to deployment.
• Evaluate and integrate commercial robotic platforms, 3D scanning systems, and sensor hardware for physical AI on the shop floor.
• Instrument robotic cells, collect real-world data, and iterate on model performance against metrics.
• Collaborate with automation and software teams to integrate solutions into the deployment architecture.

Skills

Deep learning

Computer vision

Robotics deployment

Algorithms implementation

Education

PhD required

Tools

Robotic platforms

LiDAR

Cameras

Sensor fusion

Job description

Amazon Science is seeking an Applied Scientist to create first-of-its-kind physical AI for the OMHS Software Controls and Science team. You will turn applied research into prototypes and production systems that operate on real parts on shop floors, combining ML, 3D perception, and geometric reasoning to match parts to CAD models and determine tooling without manual programming.

You will own the full loop from sensor setups to model training and production integration, enabling a lossless digital