Senior RL Scientist: Safe Legged Locomotion

Amazon; Pasadena (CA); On-site; USD 167,000 - 226,000; Full time

Job summary

Amazon Robotics, Compass Team is seeking an Applied Scientist to advance safe legged locomotion on quadraped and humanoid platforms. You will develop reinforcement learning-based controllers, integrate with model-based whole-body controllers, and push sim-to-real transfer pipelines to bridge the reality gap across varied terrains.

Collaboration with perception, planning, and safety groups will help ensure robust and safe deployment.

Qualifications

• PhD or Master’s with 6+ years of applied research experience
• 3+ years of applying RL to physical robotic systems
• Demonstrated expertise in sim-to-real transfer for locomotion or manipulation tasks
• Strong understanding of legged robot dynamics, contact mechanics, and whole-body control fundamentals

Responsibilities

• Design, train, and deploy reinforcement learning policies for dynamic legged locomotion on physical platforms
• Collaborate with safety teams to enforce safety constraints during training or deployment
• Develop sim-to-real transfer pipelines with domain randomization and system identification
• Integrate learned policies with model-based whole-body controllers and optimization layers
• Formulate reward functions encoding performance and safety constraints
• Develop and maintain large-scale RL training infrastructure and environments
• Evaluate policy performance via simulation and hardware experiments
• Publish research and contribute to Amazon's robotics leadership
• Collaborate with perception and planning for terrain-aware locomotion

Skills

Reinforcement Learning

Python

Deep Learning

Robotics

Safety-constrained RL

Control theory

Education

PhD in a relevant field

Master's degree in a relevant field

Tools

PyTorch

MuJoCo

PyBullet

Isaac Gym

JAX

Job description

Amazon Robotics, Compass Team is seeking an Applied Scientist to advance safe legged locomotion on quadraped and humanoid platforms. You will develop reinforcement learning-based controllers, integrate with model-based whole-body controllers, and push sim-to-real transfer pipelines to bridge the reality gap across varied terrains.

Collaboration with perception, planning, and safety groups will help ensure robust and safe deployment.