I see it thanks! Will take a more detailed look when I have time
1 Like
Thanks so much for this nice Plugin. I hope this will get an official, non experimental some day since it has to much potential for classic unreal AI development.
1 Like
I tried to implement this to 5.6 too - car tries to move gears and throttle just fluctuate and not much more happens. any tips or guidance if you have the updated guide appareciated! 
Hello! I ran into an issue on the course, the blueprints are not rendered fully. When I try to copy them with the ‘Copy full snippet’ button, only rendered part gets copied, which is about 20% of the full blueprint. Tried in multiple browsers on multiple devices, same issue everywhere. Also applies for the 5.3 and 5.4 version of the course. Any idea what to do?
1 Like
Hi, I am having the same issue as well with the blueprint references in the documentation not rendering properly / missing most of the nodes.
Looking into this. Thanks for reporting!
1 Like
I have the same issue in the “Design a Puzzle Adventure” tutorial, on the “Project Setup and Level Blockout” page. Trying to copy the blockout using the “Copy full snippet” buttons only copies the first 16k of text, whether or not the snippet is expanded. The problem occurs in several different browsers.
Thanks for your patience! This has been fixed.
1 Like
Thank you for the fix! I have everything working, I’m experimenting with the overtakes now. Great course and great plugin!
1 Like
Thank you for this plugin. It helped me a lot to understand the basics of RL. I documented my findings in my blog. You can find the last one here. Let me know if you think there is something wrong or missing.
1 Like
@Deathcalibur I love this tutorial thank you for making it.
I believe in 5.7.4 the part in regards to imitation learning I believe is broken. The “Make Imitation Trainer” now has a new input “Communicator”. I’m not sure how to use it properly and I don’t know if it needs new settings, it’s some kind of a struct.
Any advice would be great, as that part of the entire course is the most relevant for me 
. Anyway thanks again, hope to see a reply.
I had to change it like this, you can see the commented out parts from 5.5.
// TrainerProcess = MakeUnique(
// ULearningAgentsCommunicatorLibrary::SpawnSharedMemoryTrainingProcess(TrainerConfig->TrainerProcessSettings));
// Communicator = MakeUnique(
// ULearningAgentsCommunicatorLibrary::MakeSharedMemoryCommunicator(*TrainerProcess));
FLearningAgentsSharedMemoryCommunicatorSettings CommSettings;
TrainerProcess = ULearningAgentsCommunicatorLibrary::SpawnSharedMemoryTrainingProcess(
TrainerConfig->TrainerProcessSettings);
Communicator = ULearningAgentsCommunicatorLibrary::MakeSharedMemoryCommunicator(
TrainerProcess, TrainerConfig->TrainerProcessSettings, CommSettings);
PPOTrainer = ULearningAgentsPPOTrainer::MakePPOTrainer(
LearningAgentsManager,
Interactor,
TrainingEnv,
Policy,
Critic,
Communicator,
ULearningAgentsPPOTrainer::StaticClass(),
TEXT(“PPOTrainer”));
Looks like the 5.7 overwrites snapshots when saving. The previous one (5.5) saved them nicely with step count labels. Does anyone know if there is a way to enable it in code? I couldn’t find anything in TrainingConfig.
Do you plan to update this tutorial? Especially to include variations beyond just following a spline, and crucially to address all the node updates? Most of the nodes are obsolete now, and it’s hard to tell which are the old systems, which are the new ones, which are actually useful, and so on. :
Most stuff isn’t really obsolete (?). We just re-did how normalization and stuff is done (node names are the same), but yeah I realize it doesn’t inspire confidence to see the tutorial in shambles.
I’m pretty deep in a project right now, so I don’t have much time to update the tutorial… The good news is that we are using Learning Agents at Epic for the past two years in production and haven’t needed to make substantial rewrites of the LA API.
Thanks for the reply; I wasn’t expecting such a quick response. While the 5.4 car tutorials do explain things quite well (and with a bit of logic, you can figure out how to add other variables), the agents still weren’t picking up the observations. It turned out to be a bug (I assume): simply unplugging and replugging the “Make” pins on the “Make Shared Memory” node (for the communicator) made it work like magic. Now I can finally train my ships and try to tie everything together using the traditional behavior tree! (As for the clarity of the tutorials, I think the fact that I’m a student plays a role there, too.)
1 Like