Applied Scientist, Amazon Robotics, Compass Team

Amazon Amazon · Big Tech · Pasadena, CA · Applied Science

Applied Scientist role focused on developing and deploying AI-driven manipulation algorithms for robots in unstructured environments, with a strong emphasis on safety, contact-rich tasks, and sim-to-real transfer. The role involves designing learning-based and model-based approaches, integrating with safety software, and deploying on physical hardware.

What you'd actually do

  1. Develop and deploy manipulation algorithms for contact-rich tasks and placement across diverse object geometries and material properties
  2. Design force-controlled manipulation strategies that operate safely within Amazon Compass safety constraints
  3. Build reactive manipulation policies that detect and recover from failures (slips, missed grasps, unexpected contacts) in real time
  4. Develop learning-based manipulation policies using RL, imitation learning, or hybrid approaches, and transfer them from simulation to physical hardware
  5. Define and maintain the interface contract between manipulation algorithms and the Compass safety layer, ensuring that grasp and motion plans respect safety bounds without unnecessary conservatism

Skills

Required

  • PhD, or Master's degree and 4+ years of CS, CE, ML or related field experience
  • Experience developing manipulation algorithms that have been tested on physical robot hardware
  • Strong understanding of contact mechanics, force control, and grasp planning fundamentals
  • Proficiency in Python and C++ with experience in robotics software development
  • Experience with at least one of: reinforcement learning for manipulation, imitation learning, or model-based grasp planning
  • Familiarity with physics simulators for contact-rich tasks (e.g., Isaac Gym/Sim, MuJoCo, Drake)
  • Publication record at relevant venues (e.g., ICRA, IROS, RSS, CoRL, RA-L)

Nice to have

  • Experience in professional software development
  • Knowledge of safety-critical control (control barrier functions, constrained optimization) as it applies to manipulation
  • Experience with dexterous or multi-fingered manipulation and in-hand object reorientation
  • Experience with sim-to-real transfer for contact-rich manipulation tasks
  • Experience with compliant/impedance control and variable-stiffness actuation
  • Familiarity with foundation models or large-scale pre-training applied to manipulation
  • Experience working with multiple manipulator platforms (industrial arms, collaborative robots, custom end-effectors)
  • Exposure to functional safety concepts as they relate to human-robot interaction during manipulation
  • Strong collaboration skills and experience working in cross-functional robotics teams

What the JD emphasized

  • tested on physical robot hardware
  • contact mechanics
  • force control
  • grasp planning
  • reinforcement learning for manipulation
  • imitation learning
  • model-based grasp planning
  • physics simulators for contact-rich tasks
  • publication record

Other signals

  • Develop and deploy manipulation algorithms for contact-rich tasks and placement across diverse object geometries and material properties
  • Design force-controlled manipulation strategies that operate safely within Amazon Compass safety constraints
  • Build reactive manipulation policies that detect and recover from failures (slips, missed grasps, unexpected contacts) in real time
  • Develop learning-based manipulation policies using RL, imitation learning, or hybrid approaches, and transfer them from simulation to physical hardware
  • Define and maintain the interface contract between manipulation algorithms and the Compass safety layer, ensuring that grasp and motion plans respect safety bounds without unnecessary conservatism
  • Collaborate with perception teams to leverage object pose estimation, tactile sensing, and contact detection for closed-loop manipulation
  • Design simulation environments and training curricula for manipulation policy learning, including realistic contact physics and object diversity
  • Evaluate manipulation performance through systematic hardware experiments, measuring grasp success rates, cycle times, and safety compliance
  • Contribute to scientific publications and internal technical documentation
  • Participate in cross-team design reviews and contribute to the broader manipulation and safety architecture