Theseus: A Library for Differentiable Nonlinear Optimization
This provides a common framework for end-to-end structured learning in robotics and vision, addressing a bottleneck for researchers and practitioners in these fields, though it is incremental as it builds on existing DNLS concepts.
The authors tackled the problem of lacking an efficient, application-agnostic library for differentiable nonlinear least squares optimization by introducing Theseus, a PyTorch-based open-source library that demonstrated significant efficiency gains and better scalability in performance evaluations.
We present Theseus, an efficient application-agnostic open source library for differentiable nonlinear least squares (DNLS) optimization built on PyTorch, providing a common framework for end-to-end structured learning in robotics and vision. Existing DNLS implementations are application specific and do not always incorporate many ingredients important for efficiency. Theseus is application-agnostic, as we illustrate with several example applications that are built using the same underlying differentiable components, such as second-order optimizers, standard costs functions, and Lie groups. For efficiency, Theseus incorporates support for sparse solvers, automatic vectorization, batching, GPU acceleration, and gradient computation with implicit differentiation and direct loss minimization. We do extensive performance evaluation in a set of applications, demonstrating significant efficiency gains and better scalability when these features are incorporated. Project page: https://sites.google.com/view/theseus-ai