Compass

We are seeking a Principle Applied Scientist to join Compass. In this role, you will own the perception input into the Compass safety system, defining how robots perceive, interpret, and anticipate their surroundings in safety-critical contexts. You will develop novel approaches to environment understanding that go beyond static scene representation, providing real-time, predictive models of how humans, objects, and dynamic obstacles may evolve over short time horizons.

Your work will directly unlock robot performance by replacing conservative assumptions with precise, learned understandings of risk. You will set the scientific direction for perception within Compass, collaborate closely with controls, planning, and firmware teams, and influence the broader Amazon Robotics safety architecture.
Key job responsibilities
Define and drive the long-term scientific vision for safety-critical perception within Compass, spanning multiple robot platforms and deployment environments
Develop novel perception algorithms that provide real-time, predictive representations of dynamic environments including human motion forecasting, obstacle trajectory prediction, and scene evolution modeling
Design perception outputs that are tightly coupled to safety constraints, enabling control barrier functions to operate with minimal conservatism while maintaining formal safety guarantees
Research and develop methods to quantify and bound perception uncertainty, providing calibrated confidence estimates that safety systems can reason over
Architect perception pipelines that generalize across sensor modalities (LiDAR, depth cameras, RGB, radar) and robot morphologies without platform-specific retraining
Investigate the application of foundation models and large-scale pre-training to safety-critical perception tasks, establishing when and how learned representations can be trusted at safety-critical confidence levels
Collaborate with controls, motion planning, and firmware teams to define interface contracts between perception and downstream safety modules
Publish research at top-tier venues and represent Amazon Robotics in the broader academic and industry community
Mentor and develop a team of applied scientists and research engineers
Influence Amazon Robotics' safety architecture and perception strategy at the organizational level
About the team
Work with the inventors of control barrier functions on a novel, universal approach to safe autonomy.

You'll push the boundary of safe performance by integrating safety with motion planning, RL, and foundation models, ensuring that safety is never a blocker to robot performance.