I am using the IsaacLabExtensionTemplate as a guide for my own project, but I found something I don’t understand. The example logs various quantities for visualization using Tensorboard, amongst which is the ‘mean_reward’, but I’m unsure exactly what this is. Is it the average reward per step or per episode? Where in the code is it logging this quantity? I tried a few keywords to try to find the lines logging information on Tensorboard inside the IsaacLab repository with no luck, and debugging to try to find it was tedious and just as fruitful. Does anyone know where I can find this? Thanks.
Well, I found it, so I’m gonna leave this here in case anyone has the same doubt. The logging is performed inside the RL library, and since I was using rsl_rl, this was done inside the file on_policy_runner.py. If I understand the code correctly, mean_reward is the average of the total episode reward for the last 100 episodes to have finished.
Thanks for posting this. Our apologies we couldn’t reach your question earlier. For future reference, please post questions, proposals and bug reports in the Issues tab of Isaac Lab’s github site here.