IsaacSim 2021.2.1 RL example

Omniverse Isaac Sim
1

Hi

I’m trying to follow a standalone example of Reinforcement Learning, but I faced a trouble issue.

I just launched the python files as the document says, but the result are different to the document.

the below figures show the train parameters, and I think the RL is not executed properly.

I let it learned 5 times at the two different workstation environments, but the all results act like this video.

reward

train_para
train_para1273×816 141 KB

How can I fixed this issue, and let my agent learned successfully ?

Thank you !

2

Hi swimpark,

Are you using the same policy params and PPO params we provide in the training scripts? Few things you can try are:

  1. run the training for longer (i.e increase the total timesteps in the training script)
  2. remove goal randomization as an initial step and see if the policy learns to go to one goal only. (in the reset function in env.py)
  3. change the network architecture

Let us know if you have further questions.