Applied Scientist, Safe Control, Amazon Robotics, Compass
About the role
We are seeking an Applied Scientist to join Compass. In this role, you will develop the core Control Barrier Function (CBF) algorithms that form the mathematical foundation of the Compass safety system. You will ensure they don't just work in theory but perform reliably on real robots under real-world conditions. You will push the boundaries of concepts central to CBFs: computing robust invariant sets, designing hybrid system formulations that handle contact transitions and mode switches, and developing backup-set approaches that leverage learned policies and multiple controllers.
Responsibilities
Develop and implement novel CBF algorithms that provide formal safety guarantees while minimizing conservatism to maximize the permissible operating envelope for each robot platform
Compute and refine invariant sets for complex, high-dimensional robotic systems, developing scalable methods that go beyond what existing analytical approaches can handle
Design formulations for hybrid dynamical systems, handling discrete mode transitions (e.g., contact/no-contact, stance/flight phases) with provable safety across switching boundaries
Address the theory-to-practice gap by developing methods that are robust to model uncertainty, sensor noise, actuation delays, and computational latency
Create reduced-order and full-order dynamics models with both white-box and black-box approaches
Implement real-time optimization solvers that execute within the tight timing budgets of safety-critical control loops
Create formal arguments and documentation sufficient to support third-party safety certification of algorithms
Validate algorithms through rigorous simulation and hardware experiments, characterizing failure modes and quantifying safety margins
Contribute to the theoretical foundations of CBFs through publications at top-tier controls and robotics venues
Collaborate with perception, planning, locomotion, and manipulation teams to accommodate the needs of upstream and downstream systems
Qualifications
PhD, or Master's degree and 4+ years of deep learning, computer vision, human robotic interaction, algorithms implementation experience
Deep expertise in Control Barrier Functions, including theoretical foundations and practical implementation
Strong mathematical background in dynamical systems theory, nonlinear control, and formal verification or reachability analysis
Proficiency in C++ and Python with experience implementing control algorithms for real-time systems
Publishation record at relevant venues (e.g., CDC, ACC, ICRA, RSS, Automatica, TAC)