Skip to content

Learning in Robotics Project on multi-agent reinforcement learning for foraging on the Neural MMO platform.

Notifications You must be signed in to change notification settings

harshg99/multi_agent_foraging

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

34 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Collaborative resource sharing with multiple foraging agents

Abstract

we focus on the multi agent foraging task, where a group of agents collect resources and deposit them at a central depo in a collaborative manner. Existing decentralised solutions to foraging problem are often rule based and don’t take into account multiple objectives of exploration and exploitation for foraging, or resource constraints in agents. Our paper’s contributions are to apply a reinforcement learning solution to the multi agent foraging problem, which gives us the advantage of making a NP-hard foraging problem computationally feasible,decentralised and scalable to any number of agents under resource constraints.

Introduction

Agents would be deployed in an environment with food and water as shown in Figure 1 and each agent will collect and deposit resources while consuming a few resources to survive. This paper aims to investigate the emergence of behaviour that balances between exploitation of resources for individual survival and the team objective of depositing resources at a Depo through Reinforcement Learning. Learning for Multi agent systems can be difficult to stabilise due to non-stationarity[4] in the environment and as such we apply newer techniques in Reinforcement Learning, specifically Proximal Policy Optimisation (PPO) to train a policy. We design and compare learned policies over two reward structures for the problem.

Environment

image

Network Architecture

image

Results

image

Contributions

Harsh Goel: Environment modifications(setup,generation, resource sharing and agent observations) ,coding up and training the models, result compilations, full-report writing

Gaurav Kuppa: Environment setup, proofreading

Aditya Pratap Singh: Environment resource sharing Debugging, proofreading and model illustration

About

Learning in Robotics Project on multi-agent reinforcement learning for foraging on the Neural MMO platform.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • ASP.NET 34.2%
  • C# 24.5%
  • ShaderLab 15.1%
  • Python 12.4%
  • HLSL 8.5%
  • JavaScript 3.6%
  • Other 1.7%