search

a-n-onymous / schrl

Constrained Hierarchical Deep Reinforcement Learning with Differentiable Formal Specifications
6 stars 2 forks source link

Consultation about More Details of SCHRL #1

Open Charlie0257 opened 2 years ago

Charlie0257 commented 2 years ago

After reading your ICLR_2023 paper, Constrained Hierarchical Deep Reinforcement Learning with Differentiable Formal Specifications, I am very interested in it!

However, this project only shows part of the code so far:( I am very interested in Neural ODE which plays a role in high-level planning. (There are a lot of details that I couldn't understand when reading the article :(

Could you publish the full code?

Thanks for any help :)

a-n-onymous commented 2 years ago

Thanks for your interests in our paper!

I am on traveling these days, and we hope to publish a high-quality code repository that is easy to use, so this can take a bit longer. The pretrained models and hierarchical learning code should be public these two days.

As for neural ODE, the architecture is already public here. In the paper, Appendix F and Fig. 10(b) provide some more details about the neural ODE. We used the simplest Euler ODE with fixed time step. The key idea of neural ODE is to predict the derivative of states, instead of directly predicting the states (which is similar to ResNet).

Charlie0257 commented 2 years ago

Thank you for your answer :)

I see that you have published the code for the task. Could you update the corresponding command line in the README so that I can reproduce the results in the paper?