Sr Applied Scientist - Robotics Simulation, Amazon Robotics R&D

About the role

The Senior Applied Scientist will lead the Physical AI program at Amazon Robotics, combining deep robotics expertise with modern AI techniques. They will mentor junior scientists, architect policy training pipelines, and drive sim-to-real transfer strategies. This role involves working directly with real robot hardware, simulation environments, and production deployment pipelines.

Responsibilities

• Provide technical robotics direction for the team's Physical AI program, spanning simulation environment design, policy training, sim-to-real transfer, and real-world validation across multiple robotics platforms.
• Mentor junior applied scientists and engineers on robot learning best practices, helping them diagnose sim-to-real gaps, debug policy failures on hardware, and iterate toward deployable solutions.
• Design and execute sim-to-real transfer strategies, including system identification, domain randomization, physics parameter tuning, and visual domain adaptation, drawing on both classical and learned approaches.
• Arcitect policy training pipelines that combine teleoperation data, synthetic demonstrations, reinforcement learning, and imitation learning (e.g., VLA models, diffusion policies, behavior cloning) for manipulation tasks.
• Lead sim-to-real analysis: define metrics and methodologies for evaluating simulation fidelity, identifying where simulation diverges from reality, and prioritizing modeling improvements that impact downstream policy performance.
• Collaborate with hardware teams on robot embodiment modeling, ensuring that digital twins accurately capture kinematics, joint dynamics, actuator limits, contact behavior, and sensor characteristics.
• Evaluate and integrate state-of-the-art approaches from the Physical AI research community (foundation models for robotics, world models, action-chunking transformers, generalist policies) into the team's simulation and training infrastructure.
• Contribute to end-effector modeling and physics tuning, ensuring physically plausible contact interactions and accurate tool behavior in simulation across diverse manipulation hardware.
• Drive technical design reviews, author high-level design documents, and set the scientific direction for simulation fidelity and robot learning initiatives.

Qualifications

• PhD, or Master's degree and 6+ years of applied research experience
• Experience programming in Java, C++, Python or related language
• Experience with neural deep learning methods and machine learning
• Broad experience across a range of physics simulators (IsaacSim, IssacLab, MuJoCo, Drake, etc.), both as a sim “power-user” and technical developer
• First-hand experience in sim2real transfer (i.e. developing learned policies in sim and successfully getting them to work on real robots)
• Experiencing in closing sim2real gaps both in terms of visual fidelity and physics fidelity
• Deep expertise in robotics (controls, motion planning, perception, etc.), ideally both in the context of manipulation and locomotion
• Deep expertise in reinforcement learning, especially in the context of robotics
• Experience with VLAs and using simulation for data generation and benchmarking
• Experience with ROS2