Senior Applied Scientist, Amazon Robotics - Vulcan

Amazon Amazon · Big Tech · Seattle, WA · Machine Learning Science

Senior Applied Scientist role focused on developing and deploying learned controllers and manipulation behaviors for robotic systems in Amazon fulfillment centers, utilizing reinforcement learning and behavior cloning. The role involves research, design, implementation, and production deployment of ML models, with a strong emphasis on sim-to-real transfer and leveraging operational data for continuous improvement. The scientist will also mentor junior team members and contribute to publications.

What you'd actually do

  1. Drive the development of learned controllers and manipulation behaviors, from research prototyping through deployment on production robots.
  2. Research, design, and implement motion planning, control, and decision-making algorithms that improve the performance of deployed systems.
  3. Design and deploy learning pipelines that take policies from simulation training to reliable, real-time execution on physical robots.
  4. Develop models that predict manipulation outcomes and inform behavior selection under uncertainty.
  5. Own the development of scalable, real-time implementations — writing production-quality code and setting engineering standards for the team.

Skills

Required

  • PhD, or Master's degree and 6+ years of applied research experience
  • Experience programming in Java, C++, Python or related language
  • 4+ years of robotics work cells and control systems experience
  • 3+ years of building and deploying machine learning models for robotics applications
  • Track record of training reinforcement learning or imitation learning policies in simulation and successfully transferring them to physical robotic systems

Nice to have

  • Experience with sim-to-real transfer at scale (domain randomization, system identification, etc.)
  • Experience deploying learned policies in production environments with uptime and reliability requirements
  • Experience designing reward functions and training curricula for reinforcement learning on robotic systems
  • Publications in top robotics or machine learning venues (RSS, CoRL, ICRA, NeurIPS, ICML, etc.)

What the JD emphasized

  • robotic manipulation
  • reinforcement learning
  • behavior cloning
  • simulation alone is insufficient
  • learned approaches should replace engineered solutions
  • estimated risk
  • production robots
  • physical robots
  • real-time execution
  • production-quality code
  • real-time implementations
  • training reinforcement learning or imitation learning policies in simulation and successfully transferring them to physical robotic systems

Other signals

  • develops learned controllers and manipulation behaviors
  • applies reinforcement learning and behavior cloning to robots
  • learn from real-world data at scale
  • drive the development of learned controllers and manipulation behaviors, from research prototyping through deployment on production robots
  • design and deploy learning pipelines that take policies from simulation training to reliable, real-time execution on physical robots