search

nsidn98 / InforMARL

Code for our paper: Scalable Multi-Agent Reinforcement Learning through Intelligent Information Aggregation
https://nsidn98.github.io/InforMARL/
MIT License
91 stars 22 forks source link

Local vs Global Information #8

Closed Michael-Elrod-dev closed 8 months ago

Michael-Elrod-dev commented 8 months ago

Hello again, I wanted to ask about the node observations within this project. I see that although the agents are only meant to have access to the information within their visual radius (and their neighboring agents), they do have access to the location of their goal.

My question is, do they obtain the position of their goals through local observations or are they given this piece of global information from the beginning of each episode?

I am specifically referring to the _get_entity_feat_relative function in navigation_graph.py where the agents calculate the distance to their goal positions here: https://github.com/nsidn98/InforMARL/blob/304e905d05b34d9bf06046eb7e03904b97a14231/multiagent/custom_scenarios/navigation_graph.py#L508

nsidn98 commented 8 months ago

The agents know the global positions of their goals and also their location. This was motivated by the drone delivery systems where the drones know their target locations. I hope this answers your question!