This repository implements multiple Deep-RL agents to solve an 'openai-gym' puzzle. The idea is to share agent learned parameters to solve the puzzle ASAP!
License: MIT License
Python 100.00%
cooperative-deep-rl-multi-agents's Introduction
Solving openai-gym Puzzles using Cooperative Deep RL Multi-Agents
This repository implements multiple Deep-RL agents to solve an 'openai-gym' puzzle. The idea is to share agent learned parameters to solve the puzzle ASAP!
Strategy
Pool & share all the agent's Experience Replay Buffer.
Transfer the best agent's with other agents every episode.
All the agent's are TD3 & initialized with same parameters.
Single Agent
Training Profile
Multiple Agent
'agent1' Training Profile
'agent2' Training Profile
'agent3' Training Profile
Collective Result Analysis
Collective Training Profile
Solo & Team Agent Testing Profile
Results
Single Agent Sum.Score = -22712.6309
Multi Agent Sum.Score = -10695.2879
Single Agent Mean Score = -227.1263
Multi Agent Mean Score = -106.9528
Scores in Ratio :
Single Agent : Multi Agent = 0.4708
Multi Agent : Single Agent = 2.1236
** 3Nos. of agents in a team are 2.2136 times better than a single agent.