Real-time Learning and Planning in Environments with Swarms:A Hierarchical and a Parameter-based Simulation Approach [1]

Introduction

Project extends Craig Reynolds boid model in application of infiltration game where one attacker tries to get to the target which is defended by patrolling boids. In the codebase users can try to run it using DSS approach which is Monte Carlo Simulations. Users can also run Hierarchical approach using ARIMA model to fit ellipse over the swarm. This HAS approach will need a rJava to connect R to Java (http://rforge.net/rJava/). Naive is our baseline algorithm to compare to our DSS and HAS approaches. It is attraction vector always going in a straight line to the target and getting repulsive force from defenders if attacker happens to be too close to them.

Running the simulation

In order to run the code Adaptive Flock.jar needs to included as a module of the project as well as args parameters needs to be passed

<staring X pos> <starting Y pos> <id of run> <difficulty(hard/medium/easy)> <no. of defenders> <output path>

Directories

From DSS folder you can run Monte Carlo Simulation approach
From HAS folder you can run ARIMA model
From Naive folder you can see our baseline algorithm

References

[1] Lukasz Pelcner, Shaling Li, Matheus Aparecido do Carmo Alves, Leandro Soriano Marcolino, Alex Collins. 2020. Real-time Learning and Planning in Environments with Swarms: A Hierarchical and a Parameter-based Simulation Approach. In Proceedings of the 19th International Conference on Autonomous Agents and Multi-agents Systems (AAMAS’2020)