Senior Applied Scientist, Safe Locomotion, Compass

Description

We are seeking an Applied Scientist to join Amazon Robotics, Compass Team. In this role, you will own the development of safe legged locomotion algorithms and their deployment on physical hardware, developing learning-based controllers that enable quadrupeds and humanoids to walk, run, and recover from disturbances with agility and robustness. You will leverage Reinforcement Learning (RL), sim-to-real transfer, and other learning-based architectures to train policies that produce stable, dynamic gaits across varied terrains and operating conditions.

These learned policies will interface with model-based control strategies to form whole-body control laws that balance performance and safety. Your work sits at the novel intersection of safety and machine learning, where these learned policies will be used in a safety-critical context for complex safety constraints like stability. You will collaborate closely with perception, planning, and safety teams to close the loop between what the robot sees, where it needs to go, and how it moves to get there safely.

This is a rare opportunity to shape how legged robots move through the world alongside people.

We are seeking an Applied Scientist to join Amazon Robotics, Compass Team. In this role, you will own the development of safe legged locomotion algorithms and their deployment on physical hardware, developing learning-based controllers that enable quadrupeds and humanoids to walk, run, and recover from disturbances with agility and robustness. You will leverage Reinforcement Learning (RL), sim-to-real transfer, and other learning-based architectures to train policies that produce stable, dynamic gaits across varied terrains and operating conditions.

These learned policies will interface with model-based control strategies to form whole-body control laws that balance performance and safety. Your work sits at the novel intersection of safety and machine learning, where these learned policies will be used in a safety-critical context for complex safety constraints like stability. You will collaborate closely with perception, planning, and safety teams to close the loop between what the robot sees, where it needs to go, and how it moves to get there safely.

This is a rare opportunity to shape how legged robots move through the world alongside people.

• Design, train, and deploy reinforcement learning policies for dynamic legged locomotion including walking, running, stair climbing, and fall recovery on physical quadruped and humanoid platforms
• Collaborate with the Compass safety team to ensure locomotion policies operate within safety-critical bounds, incorporating control barrier functions or other formal safety mechanisms as constraints during or after training
• Develop sim-to-real transfer pipelines that produce policies robust to the reality gap, including domain randomization, system identification, and adaptive strategies
• Integrate learned locomotion policies with model-based whole-body controllers, defining how RL outputs (e.g., joint targets, contact schedules) interface with optimization-based control layers
• Formulate reward functions and training curricula that encode both performance objectives and safety constraints, ensuring policies respect stability and contact-force limits
• Develop and maintain large-scale training infrastructure for locomotion policy learning, including physics simulation environments and parallelized training pipelines
• Evaluate policy performance rigorously through simulation benchmarks, hardware experiments, and failure-mode analysis
• Investigate emerging techniques (e.g., foundation models for control, diffusion policies, world models) and assess their applicability to safe…