Applied Scientist – Safe RL for Dynamic Legged Robots

Amazon Science; Pasadena (CA); On-site; USD 143,000 - 193,000; Full time

Job summary

Amazon Science's SAF Lab is seeking an Applied Scientist to advance safe reinforcement learning for legged robots. You will develop policies for walking, running, climbing, and fall recovery on real hardware with safety constraints.

You will build sim-to-real pipelines, integrate with safety filters and whole-body control, and collaborate with perception, planning, and controls teams in Pasadena. Expect publishing and broad impact.

Qualifications

• PhD in Computer Science, Robotics, Mechanical Engineering, Electrical Engineering or related field with RL focus.
• Experience applying RL to physical robotic systems and sim-to-real transfer.
• Strong understanding of legged robot dynamics and whole-body control.

Responsibilities

• Collaborate with product teams and science leaders to set a science roadmap with real robot impact.
• Design, train, and deploy RL policies for dynamic legged locomotion on physical robots.
• Develop sim-to-real transfer pipelines robust to reality gaps via domain randomization and system identification.
• Integrate control-based methods with RL, including safety constraints and safety layers.
• Develop and maintain large-scale training infrastructure for locomotion policy learning.
• Investigate distillation, foundation models, and integration with perception and planning.
• Evaluate policy performance through simulation benchmarks and hardware experiments.
• Publish research at top robotics and ML venues and contribute to Amazon's scientific reputation.

Skills

Python

Deep learning

Reinforcement learning

Robotics

Sim-to-real

Education

PhD in Computer Science, Robotics, Mechanical Engineering or Electrical Engineering (RL focus)

Tools

Isaac Gym/Sim

MuJoCo

PyBullet

Job description

Amazon Science's SAF Lab is seeking an Applied Scientist to advance safe reinforcement learning for legged robots. You will develop policies for walking, running, climbing, and fall recovery on real hardware with safety constraints.

You will build sim-to-real pipelines, integrate with safety filters and whole-body control, and collaborate with perception, planning, and controls teams in Pasadena. Expect publishing and broad impact.