1234238

Thank you for your suggestion. I use the NO_CNN network and it trains very well. It’s really magical.

> Could you please share the training curve such as episode reward or success rate? The training resutlt is much related with the network type. NO_CNN net will be easier...

> TD3 needs more time to converge. Maybe you can try to increase the random explore timesteps to 5000. Always circle means the agent trapped in the local minima. Thank...

> Do you mean a custom environment? The environment is created using UE4 with AirSim plugin to provide basic API such as UAV control. You can follow this [link](https://microsoft.github.io/AirSim/unreal_custenv/) to...

> Simple_avoid is a custom environment created by myself. You can generate it from scratch in UE4 and add the AirSim plugin. As it is a long time, I cannot...

> Hello, I changed the strategy to No_CNN, but I got an error after running it twice.  I think your parameters might need to be adjusted.

> For a customed environment, we can use UE4 file to generate a good 2D background image. However, for other environment already built, such as zhangjiajie and NH, you can...