Max Robotics — Global Service Robot Directory + US Certification

About

Brandon Amos is a leading researcher at the intersection of machine learning, robotics, and differentiable optimization. His work focuses on enabling end-to-end structured learning by bridging the gap between classical optimization algorithms and modern deep learning frameworks. Amos is best known for developing **Theseus**, an open-source library for differentiable nonlinear least squares optimization built on PyTorch. This work, which has garnered 47 citations, provides a unified, application-agnostic framework for robotics and vision tasks, making it easier to integrate optimization layers into neural networks. In earlier work, Amos explored **learning awareness models**, showing that agents can learn to represent external objects purely from proprioceptive feedback—a finding with 13 citations that challenges conventional assumptions about perception and embodiment. His contributions have significant implications for robot manipulation, autonomous navigation, and structured prediction. By making optimization differentiable and accessible, Amos has helped democratize advanced techniques for end-to-end learning, influencing both academic research and practical deployment in embodied AI systems.