Supported platform about jetson-reinforcement HOT 7 OPEN

@silent-code I do plan on updating it this fall as part of an EDU course we are working on. Sorry for the delay.

from jetson-reinforcement.

What about Xavier NX?

from jetson-reinforcement.

Actually, is it possible to run this inside the ML docker?

from jetson-reinforcement.

from jetson-reinforcement.

Thanks! I'll look forward to running it!

From other comments, DRL seems to be good for the higher end platform, and the Xavier for speed. Is there anything hardware specific that will make this repo not work on the nano?

I got fairly far trying to build the repo but still encounter build errors. I manage to get through many more libraries and packages. I think I might be hitting problems with old version of code. Python 3 with Pytorch v0.9.0 seems to be working. I am still confused the issue with pip and pip3 and getting Python 2 to have Pytorch. Gym seems to be fine on both, but running the *.py files in the repo gives me graphics issues.

I have a few robotics ideas/project and the nano or the 2GB nano are really good, if I can use them.

Will there be more examples for the Xavier NX also? That is replacing the TX1 and TX2. It looks like I will run into the same "older code" issues on it. (I am borrowing one from a colleague, but have 2nd thoughts about it after your reply.)

Thanks Dusty!

from jetson-reinforcement.

Is there anything hardware specific that will make this repo not work on the nano?

There isn't anything hardware-specific that would preclude running DRL on the Nano. It's just that the Nano and Xavier NX both use a newer JetPack version than this repo supports (because the PyTorch code in this repo uses old PyTorch 0.3 and hasn't been updated yet)

from jetson-reinforcement.

The code uses older PyTorch 0.3 and needs updating. This summer I will be updating the repo to use a more modern solution. For now I recommend checking out 'Stable Baselines' repo or the repos listed at the top of the readme.
…

Dusty,

Thanks for creating this repo. You do amazing work. I feel it would serve the nvidia community well to have a self-contained RL package that works with the jetson embedded GPUs in analogy to nvidia's trt-pose and jetson-inference code bases which are fine-tunable to the user's application. Do you still plan on updating this repo for newer jetpack and pytorch versions? If so, if you could also provide examples where it is applied to a physical robot, that would be extremely helpful as there currently does not exist a stand alone, open source python RL package that doesn't require Isaak or ROS and has examples working on real mobile robots.

Thank you santa! Hehe

from jetson-reinforcement.