Question about your SA-DQN

[abderraouf2che]

I want to try your SA-DQN for example in the case of making the agent more robust to sensor malfunctions, noise and missing inputs
what's the estimated release for the code?

thanks

[huanzhang12]

Thank you for your interests in our paper!

Yes, SA-DQN can make the agent more robust to sensor malfunctions and noise. I am not sure about missing inputs though, since it can depend how the feature is represented when it is missing. If the feature becomes NaN then the whole neural network can mess up (NaN propagates through computations).

We plan to release our code very soon, most likely in two weeks. In addition, we have just updated our pre-print with more experiments as well as extending our results to PPO. It will be announced on arxiv later today.

[huanzhang12]

Our reference implementation on SA-DQN has been release: https://github.com/chenhongge/SA_DQN