Applied Scientist — Real-World Robot Manipulation

Socket.dev; Seattle (WA); On-site; USD 143,000 - 193,000; Full time

Benefits offered by this job

Medical, dental, vision coverage

401(k) plan

Paid time off

Job summary

Amazon Robotics seeks an Applied Scientist to bring learned manipulation policies to life on real robots. You will deploy, evaluate, and optimize non-prehensile behaviors in diverse, real-world tasks and contribute to scalable, real-time robot control.

You will bridge simulation and hardware, advance tactile sensing and contact-aware control, and publish results with cross-functional teams across control, perception, and hardware.

Qualifications

• PhD or Master's degree with 4+ years of science/engineering experience.
• Hands-on experience deploying learning-based control on real robots.
• Publications or patents in top venues are a plus.

Responsibilities

• Deploy and evaluate learned manipulation policies on physical robots.
• Diagnose gaps between simulated and real behavior and improve policies accordingly.
• Develop tactile, force, and contact-aware manipulation methods.
• Write production-quality code and real-time robot software.
• Build data collection and evaluation loops from real-world experiments.
• Collaborate with control, perception, and hardware teams to move ideas to hardware demonstrations.
• Represent Amazon Robotics in academia through publications and talks.

Skills

Reinforcement learning

Learning-based manipulation

Robot control

Programming (Java, C++, Python)

Education

PhD or MS + 4+ years in STEM

Tools

Robotics software stack

Perception/Control frameworks

Job description

Amazon Robotics seeks an Applied Scientist to bring learned manipulation policies to life on real robots. You will deploy, evaluate, and optimize non-prehensile behaviors in diverse, real-world tasks and contribute to scalable, real-time robot control.

You will bridge simulation and hardware, advance tactile sensing and contact-aware control, and publish results with cross-functional teams across control, perception, and hardware.