Understand the term Multi agents

[celestialdr4g0n]

From my perspective, “multi-agent” implies that the action vector is composed of joined actions, meaning each agent’s action depends on the actions/observations from other agents. This raises a couple of questions: Is it an improper use of terminology to call a single agent operating in multiple cloned environments “multi-agent”?
And what exactly does “true multi-agent” mean in this context? According to the documentation, true multi-agent training is exclusively supported by the skrl library. For further details, you can refer to their Multi-Agents Documentation.

[RandomOakForest]

Thanks for starting this discussion. When several application areas converge to propose solutions for an emerging field, it is not uncommon to have discrepancies in the terminology. In this case specifically, understanding the meaning of environment is important. The SKRL library refers to multiple agents in one environment. Cloning environments is a tool for RL. I would prefer to use the notation of single or multiple agents within the framework of one environment. Yet this notion may evolve in the future.