Multi-robot Exploration for Environmental Monitoring: The Resource Constrained Perspective

Cover
Multi-Robot Exploration
for Environmental
Monitoring:

The Resource Constrained Perspective
Copyright
Dedication
Preface
About the authors
Acknowledgments
Contents
List of figures
List of tables
Nomenclature
	Acronyms
	Constants
	Gaussian Process
	Notation
Part I
The curtain raiser
A taster of this book
1 Introduction
	1.1 Recapitulation of global emissions
	1.2 Emissions take a toll on nature
		1.2.1 California wildfires
		1.2.2 Tsunami
	1.3 Pollution takes a toll on human health
	1.4 Need for environment monitoring
	1.5 Conventional methods
	1.6 Book organization
	References
2 Target environment
	2.1 Aerial environments
	2.2 Marine environments
	2.3 Ground environments
	2.4 Types of observations
	2.5 Types of predictions
	2.6 Summary
	References
3 Utilizing robots
	3.1 Unmanned Aerial Vehicles (UAVs)
		3.1.1 Rotary-winged UAVs
		3.1.2 Fixed-winged UAVs
	3.2 Unmanned Marine Vehicles (UMVs)
		3.2.1 Underwater vehicles
		3.2.2 Surface vehicles
	3.3 Unmanned Ground Vehicles (UGVs)
		3.3.1 Wheels
		3.3.2 Continuous tracks
	3.4 Sensors
		3.4.1 Sensor range
			3.4.1.1 Short-range sensors
			3.4.1.2 Long-range sensors
		3.4.2 Sensory information
			3.4.2.1 Interoceptive sensors
			3.4.2.2 Exteroceptive sensors
	3.5 Real-life example
	3.6 Summary
	References
4 Simultaneous Localization and Mapping (SLAM)
	4.1 Mapping
		4.1.1 Metric maps
		4.1.2 Topological maps
		4.1.3 Topometric maps
		4.1.4 Semi-metric topological maps
		4.1.5 Measurement maps
	4.2 Localization
		4.2.1 Global localization
		4.2.2 Local localization
	4.3 Simultaneous Localization and Mapping (SLAM)
		4.3.1 Conventional (probabilistic) SLAM
		4.3.2 Bio-inspired SLAM
	4.4 Summary
	References
Part II
The essentials
The building blocks
5 Preliminaries
	5.1 Bayesian Inference (BI)
	5.2 Multi-variate Gaussian (MVN)
	5.3 Gradient Descent (GD)
		5.3.1 What is a gradient?
		5.3.2 How does it work?
		5.3.3 How optimal is the optimal solution?
		5.3.4 Types of gradient descent
			5.3.4.1 Batch Gradient Descent (BGD)
			5.3.4.2 Stochastic Gradient Descent (SGD)
			5.3.4.3 Mini-batch Gradient Descent (MBD)
		5.3.5 Gradient Descent with Random Restarts (GDR)
		5.3.6 Termination condition
	5.4 Kernel trick
	5.5 Cholesky decomposition for kernel inversion
	5.6 Using jitter for numerical stability
	5.7 Summary
	References
6 Gaussian process
	6.1 Gaussian Process (GP)
		6.1.1 Parametric versus non-parametric
		6.1.2 Gaussian distribution vs process
		6.1.3 Notational conventions
		6.1.4 Mean function
		6.1.5 Covariance function
		6.1.6 Choices of kernels
	6.2 Kernel jargons
	6.3 Bayesian inference
		6.3.1 Prior
		6.3.2 Likelihood
		6.3.3 Posterior
		6.3.4 Maximum Likelihood Estimation (MLE)
	6.4 Entropy
	6.5 Multi-output GPs (MOGPs)
	6.6 Limitations of GPs
	6.7 Approximate GPs
		6.7.1 Global approximation methods
			6.7.1.1 Random subset selection
			6.7.1.2 Active subset selection
			6.7.1.3 Sparsifying the kernel
			6.7.1.4 Sparse approximation of the kernel
			6.7.1.5 Variational sparse approximation
		6.7.2 Local approximation methods
	6.8 Applications of GPs
		6.8.1 GPs applied to classifications tasks
		6.8.2 GPs applied to regression tasks
		6.8.3 GP applied to Bayesian Optimization (BO)
	6.9 Hands-on experience with GPs
	6.10 Pitfalls
	6.11 Summary
	References
7 Coverage Path Planning (CPP)
	7.1 Coverage path planning in combinatorics
		7.1.1 Covering Salesman Problem (CSP)
		7.1.2 Piano mover\'s problem
		7.1.3 Art gallery problem
		7.1.4 Watchman route problem
		7.1.5 Orienteering Problem (OP) for vehicle routing
	7.2 Coverage path planning in robotics
		7.2.1 Lawn mower strategy
		7.2.2 Frontier based exploration
		7.2.3 Voronoi tessellation based area coverage
	7.3 Coverage path planning in Wireless Sensor Networks (WSNs)
	7.4 Challenges
	References
8 Informative Path Planning (IPP)
	8.1 Planning over waypoints
		8.1.1 Selection of waypoints
		8.1.2 Notational convention for inputs and targets
		8.1.3 Entropy maximization (full-DAS)
		8.1.4 Nearest Neighbor (NN)
		8.1.5 Resource utilization efficacy amelioration while maximizing information gain
		8.1.6 Comparative analysis of information acquisition functions
	8.2 Homing
		8.2.1 Dynamically choosing weights for optimization
		8.2.2 Additional homing constraints
	8.3 Experiments
		8.3.1 Dataset
		8.3.2 Analysis without homing guarantees
		8.3.3 Analysis with homing guarantees
			8.3.3.1 Path cost analysis with homing enforced
			8.3.3.2 Model quality analysis with homing enforced
	8.4 Summary
	References
Part III
Mission characterization
How does one define a mission?
9 Problem formulation
	9.1 Relationship between robots and GP
	9.2 Scenario
		9.2.1 Starting configuration
		9.2.2 Communication strategy
		9.2.3 Sensing
		9.2.4 Mission termination conditions
		9.2.5 Model fusion
	9.3 Summary
10 Endurance & energy estimation
	10.1 Endurance estimation: the notion
	10.2 Energy estimation: the notion
	10.3 Conclusion
	References
11 Range estimation
	11.1 Importance of Operational Range Estimation (ORangE)
	11.2 Rationale behind the maverick approach
	11.3 Workflow
	11.4 Simplified range estimation framework for UGV
		11.4.1 Energy distribution model
			11.4.1.1 System identification
		11.4.2 Simplified Operational Range Estimation (ORangE)
	11.5 Generic Range Estimation (ORangE) framework for diverse classes of robots
		11.5.1 First things first
		11.5.2 Enhancements over the simplified framework
		11.5.3 Energy distribution model for diverse robots
		11.5.4 Range estimation models for diverse robots
			11.5.4.1 Approach 1: Offline Operational Range Estimation (Offline ORangE) for diverse mobile robot platforms
				1 Case 1: UGV operating on uneven terrain
				2 Case 2: multi-rotor UAV operating in the presence of external disturbances
			11.5.4.2 Approach 2: Online Operational Range Estimation (Online ORangE) for diverse mobile robot platforms
				3 Case 1: UGV operating on uneven terrain
				4 Case 2: multi-rotor UAV operating in the presence of external disturbances
	11.6 Experiments
		11.6.1 System identification
		11.6.2 Indoor experiments
		11.6.3 Outdoor experiments
		11.6.4 Batteries used for field experiments
		11.6.5 Case 1: UGV
		11.6.6 Case 2: UAV
	11.7 Summary
	References
Part IV Scaling to multiple robots
	IV.1 Multi-robot systems
	IV.2 Fusion of information from multiple robots
12 Multi-robot systems
	12.1 Advantages of scaling
	12.2 Challenges to scaling
		12.2.1 Selecting optimal team control strategy
			12.2.1.1 Centralized strategies
			12.2.1.2 Decentralized strategies
			12.2.1.3 Distributed strategies
			12.2.1.4 Solitary confinement strategies
		12.2.2 Selecting optimal team communication strategy
			12.2.2.1 Synchronous communication
			12.2.2.2 Asynchronous communication
			12.2.2.3 Disconnected strategies
		12.2.3 Tackling rogue agents
	12.3 Summary
	References
13 Fusion of information from multiple robots
	13.1 Overview of the disconnected-decentralized teams
	13.2 Multi-robot sensing scenario
	13.3 Various notions of fusion
	13.4 Existing fusion approaches
	13.5 Limitations of existing works
	13.6 Notational conventions
	13.7 Predictive model fusion for distributed GP experts (FuDGE)
		13.7.1 Fusion strategy
			13.7.1.1 Point-wise mixture of experts using GMM
			13.7.1.2 Generalized product-of-experts model [15]
			13.7.1.3 Multiple mobile sensor nodes generating single GP
	13.8 Map-Reduce Gaussian Process (MR-GP) framework
	13.9 Experiments
		13.9.1 Fusion quality
		13.9.2 Path length
		13.9.3 Computational complexity
	13.10 Conclusion
	References
Part V Continuous spatio-temporal dynamics
	V.1 Spatio-temporal analysis
14 Spatio-temporal analysis
	14.1 From discrete to continuous space
	14.2 Spatio-temporal kernels
	14.3 Information acquisition for spatio-temporal inference
	14.4 Summary
	References
Part VI Epilogue
	VI.1 Real-world algal bloom monitoring
	VI.2 Cumulus cloud monitoring
	VI.3 Search & rescue
	VI.4 Conclusion
15 Real-world algal bloom monitoring
	15.1 Motivation
	15.2 Success with real-world deployments
	15.3 Common deployment strategies
	15.4 Deployment challenges
	15.5 Open research problems
	References
16 Cumulus cloud monitoring
	16.1 Motivation
	16.2 Challenges
	16.3 Success story
	References
17 Search & rescue
	17.1 Conventional search-and-rescue missions
	17.2 Modern search-and-rescue missions
	17.3 Future search-and-rescue missions
	17.4 Summary
	References
18 Received signal strength based localization
	18.1 Localization based on signal strength
	18.2 Challenges to received signal-strength based localization
		18.2.1 Lack of labeled training data
		18.2.2 Sparsity of training data
		18.2.3 Propagating location uncertainty while training
	18.3 Summary
	References
19 Conclusion & discussion
	19.1 Summary of contributions
	19.2 Significance of contributions
	19.3 Further works
		19.3.1 Necessary extensions
		19.3.2 Sufficient extensions
	19.4 Closing remarks
	References
List of reproduced material
	List of reproduced figures
		Chapter 1
		Chapter 2
		Chapter 3
		Chapter 4
		Chapter 7
		Chapter 8
		Chapter 10
		Chapter 11
		Chapter 12
		Chapter 13
		Chapter 15
		Chapter 16
		Chapter 17
		Chapter 18
	List of reproduced tables
		Chapter 8
		Chapter 11
		Chapter 13
Index
Back Cover