STLCG++: A New and Improved Signal Temporal Logic Toolbox in Jax and PyTorch!
We have just released STLCG++, a new and updated version of STLCG which leverages automatic differentiation tools to make evaluating and differentiation STL robustness formulas efficient. The STLCG++ implementation is inspired by the transformer architecture, similar to how the original STLCG implementation was inspired by recurrent neural networks. We show that STLCG++ can be easily used for parameter mining, trajectory optimization, and diffusion modeling applications.
This work was performed in collaboration with Parv Kapoor and Eunsuk Kang.
The paper, project page, and code (Jax and PyTorch) are available online. The code repo has some demo jupyter notebooks for you to try out!
Please check it out!