Hi Soumya Maitra! Thanks for watching my video! I am using RTX2060, 500GB SSD and a i9-9900K processor. You can find required specifications for Isaac sim here. docs.omniverse.nvidia.com/isaacsim/latest/installation/requirements.html
This is an excellent tutorial! Thank you for the effort you put into it and for sharing it. I’d like to suggest covering how to use the trained policy in Isaac Sim in a future video. Do you think it would be possible to use this policy while still controlling the robot to move around in a scene? I would also be interested in getting in touch with you. Would it be possible for you to share a way to reach you? Many thanks again!
Hi Filipe Simões! Thanks for watching my video! At the very end of the video, I explained how to use trained policy (13:00~). By executing “play.py” you can use trained policy (model). I think it is possible to train the robot to make it move around, but in that case, it will take much more time and maybe another method is required. Here is my e-mail address: robotmania8867@yahoo.com
Thank you very much RobotMania for this interesting video tutorial. I wonder how difficult or straightforward it would be to use the resulting policy from this training on a simulated robot in Gazebo. It would be great if you could make a video explaining how to use a model trained in Isaac Lab in the ROS environment. The only one I have found on this subject (using the resulting policy from Isaac Lab in an environment other than Isaac Sim) is the one in the official documentation where they implement the trained algorithm on the real robot Spot from Boston Dynamics. The problem is that it seems to use a specific API for this robot, which I don't know if it is based on ROS2.
Hi Jose Luis Torres! Thanks for watching my video! Isaac sim uses SB3 API as it is, so you can use trained data in another simulator like Gazebo as long as you follow specified protocol. In case you create custom environment for SB3, you should follow these instructions. stable-baselines3.readthedocs.io/en/master/guide/custom_env.html If you would like to create custom environment for gazebo, you should create the “CustomEnv” class inside ROS2 node. It will look like this. class GazeboEnv(Node): def __init__(self): super().__init__('env') class CustomEnv(gym.Env): def __init__(self, arg1, arg2, ...):
Thank you! Your tutorials are awesome-keep going! I have a question, if you can help me: Do you have an idea how I can draw a trajectory in Isaac Lab to use it in the environment for training, for example, a robot to follow a certain trajectory?
Hi street world TV! Thanks for watching my video! It is possible to add texture to assets in the Isaac Sim (Lab). So you can draw a trajectory on the png file and map it on the object.
Awesome tutorial and content. But I do wonder if there is a tutorial on how to write the RL parameter configurations, namely that .yaml file. I cannot seem to find this in the Isaac Lab official documentations, at least not anywhere conspicuous. I do think these configuration files stem from Isaac Lab wrapped RL library interfaces, so it's odd that there's no relevant information. Do you know how to operate this, or where to find relevant information? Thanks a lot!
Hi Tony YANG! Thanks for watching my video! I would rather recommend you see “Stable Baselines3” documentation. Here is the page for PPO which I used in my tutorial. stable-baselines3.readthedocs.io/en/master/modules/ppo.html
I think you can refer this documentation isaac-sim.github.io/IsaacLab/main/source/tutorials/index.html and you can try to add small modifications to the existing code to see what happens.
When I tried to run with the same command line with you to train the robot (using sb3 and num_envs 8), I was success to run the program. However, the result is very bad compared with yours. In your case, I can see the robot can balance itself when you run the play code. In my case, the robot still fell down. Do you have any idea what would be the reason for this? Ive tried to increase the num_envs, but still cannot reach your result in the video. Thanks!
Hi Towerboi! I also find that training results are unstable. I did train in several machines and results were different each time. Please try to experiment with hyper parameters in the “sb3_ppo_cfg” file. I actually got better results with less timesteps (1e6). If you have time, you can do grid search of hyper parameters.
Hi LycheeAI! Thanks for watching my video! I am creating these videos among my daily work. So, the time when I can do it is very volatile and it is difficult to make any plans or collaborations. But I appreciate your suggestion.
@@robotmania8896 I understand, perhaps if you need help with anything in the future regarding Isaac Lab or on a video, I could help you realize it faster. Thanks for putting these videos out there, let's make robotics bigger!
Nice tutorial, very awesome job, I got stuck in the joint param settings for a long time. thanks a lot
Hi 富华 贾!
Thanks for watching my video!
It is my pleasure if this video has helped you!
Awesome tutorials. Just loved it.
Please let us know your machine configuration, I am unsure if Isac sim or training can be done in my PC. Thanks!
Hi Soumya Maitra!
Thanks for watching my video!
I am using RTX2060, 500GB SSD and a i9-9900K processor. You can find required specifications for Isaac sim here. docs.omniverse.nvidia.com/isaacsim/latest/installation/requirements.html
@ thank you so much.
Keep up the awesome work !
Thank you! A wonderful lesson!
You are welcome!
Thanks bro, you just save my day!
Hi Đạt Lê!
Thanks for watching my video!
It is my pleasure if this video has helped you.
Timely! Thank you!
Hi chevotebenadoa 4!
Thanks for watching my video!
It is my pleasure if this video has helped you.
Thankyou for this video brother
Hi Manu Mohan!
You are welcome!
This is an excellent tutorial! Thank you for the effort you put into it and for sharing it.
I’d like to suggest covering how to use the trained policy in Isaac Sim in a future video. Do you think it would be possible to use this policy while still controlling the robot to move around in a scene?
I would also be interested in getting in touch with you. Would it be possible for you to share a way to reach you?
Many thanks again!
Hi Filipe Simões!
Thanks for watching my video!
At the very end of the video, I explained how to use trained policy (13:00~). By executing “play.py” you can use trained policy (model). I think it is possible to train the robot to make it move around, but in that case, it will take much more time and maybe another method is required. Here is my e-mail address: robotmania8867@yahoo.com
Thank you very much RobotMania for this interesting video tutorial. I wonder how difficult or straightforward it would be to use the resulting policy from this training on a simulated robot in Gazebo. It would be great if you could make a video explaining how to use a model trained in Isaac Lab in the ROS environment. The only one I have found on this subject (using the resulting policy from Isaac Lab in an environment other than Isaac Sim) is the one in the official documentation where they implement the trained algorithm on the real robot Spot from Boston Dynamics. The problem is that it seems to use a specific API for this robot, which I don't know if it is based on ROS2.
Hi Jose Luis Torres!
Thanks for watching my video!
Isaac sim uses SB3 API as it is, so you can use trained data in another simulator like Gazebo as long as you follow specified protocol. In case you create custom environment for SB3, you should follow these instructions. stable-baselines3.readthedocs.io/en/master/guide/custom_env.html
If you would like to create custom environment for gazebo, you should create the “CustomEnv” class inside ROS2 node. It will look like this.
class GazeboEnv(Node):
def __init__(self):
super().__init__('env')
class CustomEnv(gym.Env):
def __init__(self, arg1, arg2, ...):
Thank you! Your tutorials are awesome-keep going! I have a question, if you can help me: Do you have an idea how I can draw a trajectory in Isaac Lab to use it in the environment for training, for example, a robot to follow a certain trajectory?
Hi street world TV!
Thanks for watching my video!
It is possible to add texture to assets in the Isaac Sim (Lab). So you can draw a trajectory on the png file and map it on the object.
Thank you for this tutorial! How much time do I need to run the training for a good resault?
Hi קרן רייף!
Thanks for watching my video!
It depends on the computational capability of your computer, but I think it will take a couple of hours.
Awesome tutorial and content. But I do wonder if there is a tutorial on how to write the RL parameter configurations, namely that .yaml file. I cannot seem to find this in the Isaac Lab official documentations, at least not anywhere conspicuous. I do think these configuration files stem from Isaac Lab wrapped RL library interfaces, so it's odd that there's no relevant information. Do you know how to operate this, or where to find relevant information? Thanks a lot!
I just found the Gymnasium documentations, could be something here, looking at it now.
Hi Tony YANG!
Thanks for watching my video!
I would rather recommend you see “Stable Baselines3” documentation. Here is the page for PPO which I used in my tutorial.
stable-baselines3.readthedocs.io/en/master/modules/ppo.html
Thank you so much for this tutorial. Even following this is pretty hard. How to learn that code of RL?
I think you can refer this documentation isaac-sim.github.io/IsaacLab/main/source/tutorials/index.html and you can try to add small modifications to the existing code to see what happens.
Hi, can you do some coppelia sim manipulation with moveit?
Hi Marco Neves!
Thanks for watching my video!
I didn’t even know that such simulator exists. Thank you for information. I will consider it.
Thank you bro!
You are welcom!
When I tried to run with the same command line with you to train the robot (using sb3 and num_envs 8), I was success to run the program. However, the result is very bad compared with yours. In your case, I can see the robot can balance itself when you run the play code. In my case, the robot still fell down. Do you have any idea what would be the reason for this? Ive tried to increase the num_envs, but still cannot reach your result in the video. Thanks!
Hi Towerboi!
I also find that training results are unstable. I did train in several machines and results were different each time. Please try to experiment with hyper parameters in the “sb3_ppo_cfg” file. I actually got better results with less timesteps (1e6). If you have time, you can do grid search of hyper parameters.
I see, thank you for your suggestion!
Awesome Video!
Been working on Isaac Lab Tutorials as well, let's collaborate if you'd like to!
Hi LycheeAI!
Thanks for watching my video!
I am creating these videos among my daily work. So, the time when I can do it is very volatile and it is difficult to make any plans or collaborations. But I appreciate your suggestion.
@@robotmania8896 I understand, perhaps if you need help with anything in the future regarding Isaac Lab or on a video, I could help you realize it faster. Thanks for putting these videos out there, let's make robotics bigger!