RayPerceptionSensorComponent3D Clarification

Hi guys..

I just wanted to ask RayPerceptionSensorComponent3D component. As far as I see from the threads, as I attach the RayPerceptionSensorComponent3D to my agent, I don't have to mention or reference or set any values right? The ML-Agents library will take the control and do the learning duty without any more settings. Is that true? If so, its like magic loved that.. A bit of control would be nice though. If so I would love to know best practice to debug this component.
Cheers and wish you happy, healthy days ahead.
Fau

 

Hi Jeffrey,
Thanks for the rapid reply. I just wanted to have more of a control. Sort of.. I mean, Unity does everything but how does the RayPerceptionSensorComponent3D component effect the performance? In terms of memory, iterations, rewards etc... Besides, to be precise vector sensors are created behind the scenes and I wanted to trace the component Also If I had an insight maybe I could find out more questions.. I have implemented the component and running a curriculum learning process right now. Yes, I could compare the results from tensor flow but if I had an opportunity via the editor, that would be cool.

A question popped into my mind just right now. What if I remove these sensor components after the model is populated. Will the agent need all the sensors during the game play that I have attaced for learning process? If yes, how will these sensors affect the performance? I wanted to analyse these points. And with that maybe, I will use these agents with DOTS or Physics.. So, the component might require a deeper investigation.

Thanks again for the "thunder" reply
Cheers

 

I've wondered about performance as well, seeing as it would be adding several raycasts to the player each frame if these are used during normal gameplay when not actually training the agent.