Offline reinforcement learning is to learn from a fixed dataset without further interactions with the environment. There are many papers focusing on offline RL for a single-agent setting. e.g. https://offline-rl-neurips.github.io/papers.html, However, no literature research on how to learn in multiagent settings. I have tried several state-of-the-art offline reinforcement learning approaches, but it doesn't work well in a multiagent environment. My code can be found here: https://github.com/SHITIANYU-hue/multiagent-offline-RL. The main challenge is how to learn multiagent distribution.
https://arxiv.org/pdf/2006.13888
http://www.dcsc.tudelft.nl/~bdeschutter/pub/rep/10_003.pdf
Article Offline reinforcement learning with task hierarchies
Aravinda C V thanks, but they haven't discussed how to solve multiagent offline RL problems
https://towardsdatascience.com/multiagent-reinforcement-learning-for-robocup-rescue-simulator-17773553f46
https://arxiv.org/pdf/1912.00498
https://rlss.inria.fr/files/2019/07/RLSS_Multiagent.pdf
- Badges
- Science topic
- Similar topics
- Computer Science
- Program