Test with action_mean in cleanrl

ntnu-arl / aerial_gym_simulator

Aerial Gym Simulator - Isaac Gym Simulator for Aerial Robots

https://ntnu-arl.github.io/aerial_gym_simulator/

BSD 3-Clause "New" or "Revised" License

353 stars 58 forks source link

Test with action_mean in cleanrl #9

Open LjXiii opened 4 months ago

LjXiii commented 4 months ago

I want to use action_mean (instead of random sampling) to test the already trained neural network model in CleanRL. The results are very poor, but using the original random sampling method for testing yields very good results. I would like to ask why this might be happening?

The testing code is modified as shown in the uploaded image, using action_mean for testing. test

mihirk284 commented 3 months ago

Hello @LjXiii, I am not sure why the specific RL framework is not yielding desired results. I had tried the same with Sample-Factory and was getting expected results. Do other frameworks also cause a similar behavior?