[Question] Scan observation space

[Paulchen29]

❓ Question

Hi

I have a really huge obs space like a big room maybe 20 by 20 meters! In this room there are blocks which my agent shall pickup and put on defined locations! I am aware of framestacking but as the blocks have details the agent needs to understand I don't want to scale down everything too much down! So my idea is to split my huge image in small 84×84 blocks (but not scaled) and I want the agent to scan and understand the obs before he starts! Like I think the vacuum cleaning bots are doing - they scan first the room and then plan the route! Is this possible with stb3 and if yes where is a good starting point of documentation to drive in! Thanks in advance

Jürgen

Checklist

• [X] I have checked that there is no similar issue in the repo
• [X] I have read the documentation
• [X] If code there is, it is minimal and working
• [X] If code there is, it is formatted using the markdown code blocks for both code and stack traces.

[qgallouedec]

It reminds me a little of https://github.com/araffin/aae-train-donkeycar. But, unfortunately, we do not do technical support. Please post your question on the RL Discord, Reddit or Stack Overflow in that case.