search

alizandian / safe-split-dqn

Safer Deep Reinforcement Learning (DRL) by iteratively updating domain variables including Safety Monitors. Currently extending work from here: https://github.com/jinwooro/Safety-RL-Generic
3 stars 0 forks source link

Papers #1

Open TrinhTuanHung2021 opened 2 years ago

TrinhTuanHung2021 commented 2 years ago

Hello. Thank you for sharing your model. Could you upload your paper or documents related to your model?

alizandian commented 2 years ago

Hello. Thank you for sharing your model. Could you upload your paper or documents related to your model?

Of course, but currently I'm working on it. I'll close this issue when I upload them.

TrinhTuanHung2021 commented 2 years ago

Hello. Thank you for sharing your model. Could you upload your paper or documents related to your model?

Of course, but currently I'm working on it. I'll close this issue when I upload them.

Thank you for your reply.

Could you guide me to run the model?

alizandian commented 2 years ago

In the main.py, I have some examples of how to use the model, very generically. The current model I'm working on is AgentIterativeSafetyGraph, which is a Reinforcement learning (q value type) Agent with a monitor, and also a safety graph, that accompanies the neural network, mainly to refine the experiences fed to the reinforcement learning. Agent will return safe actions after learning phase (Epsilon is surpassed). But again, I'm currently working on it so some parts of it are hard coded or not finished yet. Model is kinda finished, but I'll do some more refinements and add more Environments for benchmarking in the future. I'll also link my paper in next months when it was finished.