Interesting. It will be nice to see how we can now experiment with our own RL choices and implementations. a) How we can use simple RL Algorithms, like REINFORCE against these simple Unity Agent Environments. (With Unity we are stuck with PPO and SARSA) b) And then possibly some more advanced community/ library RL Algorithms c) And see how these compare to ML-Agents performance, and Training speeds. Since ML-Agents has the advantage of multiple environments, and running at faster than user speeds. (I guess you can to, since you have the simulate component)
I get better results, faster, training the DungeonEscape scenario using PeacefulPie, than with MLAgents. You don't need multiple environments, within a single Unity instance, for speed, you just need speed. Unity is mostly single-threaded, so multiple environments within a single Unity instance doesn't actually improve throughput, unless you have some arbitrary bottleneck, like running at a specific framerate.
You can, yes! The easiest way is to build a standalone server. I did this for linux (built on Mac, run on linux), and used this to run 8 instances on an aws gpu instance. Ran ok :)
![2 Create Unity RL env WITHOUT mlagents! [v1, with background music, and longer transitions]](http://i.ytimg.com/vi/p34dOZ6fcqg/mqdefault.jpg)
![2 Create Unity RL env WITHOUT mlagents! [v1, with background music, and longer transitions]](/img/tr.png)
Interesting. It will be nice to see how we can now experiment with our own RL choices and implementations.
a) How we can use simple RL Algorithms, like REINFORCE against these simple Unity Agent Environments. (With Unity we are stuck with PPO and SARSA)
b) And then possibly some more advanced community/ library RL Algorithms
c) And see how these compare to ML-Agents performance, and Training speeds. Since ML-Agents has the advantage of multiple environments, and running at faster than user speeds. (I guess you can to, since you have the simulate component)
I get better results, faster, training the DungeonEscape scenario using PeacefulPie, than with MLAgents. You don't need multiple environments, within a single Unity instance, for speed, you just need speed. Unity is mostly single-threaded, so multiple environments within a single Unity instance doesn't actually improve throughput, unless you have some arbitrary bottleneck, like running at a specific framerate.
I guess I should create a video to demonstrate this.
Thank you
I have question about env.
Can we send image from unity to Python code using peaceful-pie??
Yes, e.g by converting into a base64 string.
I don't know how to do that can you explain more?@@rlhugh
Hi
Can I run unity without rendering GUI?
You can, yes! The easiest way is to build a standalone server. I did this for linux (built on Mac, run on linux), and used this to run 8 instances on an aws gpu instance. Ran ok :)
Can you please help me with your RayCasts script i have added the script and added detectable how to get necessary info for obs ????
Brother sorry to bother you it would be a great help