Multiagent patrolling and reinforcement learning

For more info about our services contact : help@bestpfe.com

Table of contents

List of figures
List of tables
Nomenclature
1 Introduction
1.1 Context and motivations
1.2 Overview
2 State of the art
2.1 Multiagent patrolling
2.1.1 Environment
2.1.2 Society of agents
2.1.3 Performance measures
2.1.4 Discrete time model
2.2 Methodology
2.3 Classical strategies
2.3.1 Reactive strategies
2.3.2 Cognitive strategies
2.3.3 Graph-theory-based strategies
2.3.4 Markov-decision-process-based strategies
2.4 Machine-learning-based strategies
2.4.1 Bayesian learning
2.4.2 Neural-learning-based patrolling
2.5 Conclusion
3 Model, methodology and implementation
3.1 Model and definitions
3.1.1 Model of the MAP problem
3.1.2 Definitions
3.2 Types of stationary strategies and structure of resultant data
3.3 Methodology
3.4 Implementation
3.4.1 PyTrol
3.4.2 MAPTrainer
3.4.3 MAPTor
3.5 Model strategy and databases
3.5.1 First database: HCC 0.2
3.5.2 Second database: HPCC 0.5
3.6 Conclusion
4 Path-Maker: a decentralised strategy based on node prediction
4.1 Path-Maker
4.1.1 Path-Maker
4.1.2 RNN-Path-Maker: an implementation of Path-Maker
4.1.3 Deterministic-Path-Maker
4.1.4 Random-Path-Maker
4.2 Training procedure
4.2.1 Pretraining
4.2.2 Main training
4.3 Experiments and results
4.3.1 Conduct of experiments
4.3.2 Training settings
4.3.3 Training results
4.3.4 Simulation results
4.4 Conclusion
5 RAMPAGER: a strategy relying on structure-guided LSTM initialisation
5.1 Procedure of training
5.1.1 Structure-guided initialisation: an analytical initialisation
5.2 Experiments and results
5.2.1 Conduct of experiments
5.2.2 Preliminary experiments: selection of the LSTM setting
5.2.3 Training settings
5.2.4 Training results
5.2.5 Simulation results
5.3 Conclusion
6 Idleness estimator: a decentralised strategy based on idleness estimation
6.1 Idleness estimation
6.2 Decision-making based on idleness estimation
6.2.1 Deterministic approach
6.2.2 Drawback of the deterministic approach
6.2.3 Stochastic approach
6.3 Some statistical models for idleness estimation
6.4 Training procedure
6.5 Experiments and results
6.5.1 Training settings
6.5.2 Training results
6.5.3 Simulation results
6.6 Conclusion
7 Interacting Idleness Estimator: a strategy based on interaction
7.1 Interacting Idleness Estimator
7.1.1 Peer-to-peer interaction
7.1.2 Transitive interaction
7.2 Experiments and results
7.2.1 Training results on HPCC 0.5 data
7.2.2 Simulation results
7.3 Conclusion
8 Conclusion
References