search

leggedrobotics / rsl_rl

Fast and simple implementation of RL algorithms, designed to run fully on GPU.
Other
645 stars 184 forks source link

Is rsl_rl v2.0.x supporting legged_gym? #25

Open JLCucumber opened 8 months ago

JLCucumber commented 8 months ago

Hi,

I am using legged_gym currently for legged robots training. I noticed that this package rsl_rl has been updated to 2.0 version while the required version for legged_gym is v1.0.2. I wonder if it is available to adopt this 2.0 version to my legged gym without potential problems, since I find myself in need of the newly-added tensorboard to view the trainning performance.

I am new to tensorboard and rl coding, so I'm afraid manually adding a tensorboard by myself to my original code may result in some problems. Thank you for your help : )

JLCucumber commented 8 months ago

Never mind, I found the answer in my code 🤡 👈 🤣

A tensorboard tool for viewing my training results is already available in rsl_rl v1.0.2. It is located in rsl_rl/runners/on_policy_runner.py and is a bunch of code blocks with self.writer.add_scalar. After each training session, this particular file is saved to the exact location of the trained model, usually legged_gym/logs/[your task name]/[your model timestamp].

Here's how to use it (for my task):

  1. cd /home/jlcucumber/nvidia/legged_gym/logs/dogzilla/
  2. tensorboard --logdir ". /Mar0317-18-24"
  3. Then it will show the check link and hopefully you will see something like this:

! image

Cheers