-
Notifications
You must be signed in to change notification settings - Fork 16
Give me knowledge
The following sections present references, such as books, scientific papers, and online courses, that helps anyone to understand the theoretical foundations used in the project's development.
Probability theory is the core of any modern, robust technique for state estimation and machine learning algorithms. Basic probability knowledge is require to understand the fundamentals of many techniques used in this project.
- Paul L. Meyer - Probabilidade: Aplicações à Estatística: Meyer's book is a great undergraduate-level introduction to probability theory and is strongly recommended as the first place to start looking at.
- Daphne Koller - Probabilistic Graphical Models: Koller's book presents a unique way to understand and reason about probability using graphs to help visualizing what's going on. It isn't a required reading, although pretty interesting specially by watching the Coursera MOOC on PGM.
Even though only simulation has been done during the system's development, the ultimate goal it to provide algorithms suitable for use in robots. Therefore, overall knowledge in Robotics is necessary to understand the limitations and challenges of developing algorithms that will be executed in the real world.
- Maja J. Mataric - The Robotics Primer: This little book introduces Robotics in a simple language that anyone can understand. The fundamentals are multidisciplinary and basic for any further study with robots and their complexities. This book is strongly recommended for individuals who have no knowledge in the field.
- Robin R. Murphy - Introduction to AI Robotics: The book is focused in computational challenges when using Artificial Intelligence in Robotics, specially the ones in system's architecture. It presents several examples of what has been done, epic failures, and how scientists and engineerings overcame challenges throughout history in order to create better and more useful robots.
The world is stochastic, i.e., we cannot be absolutely certain about anything that is going on. Therefore, Probabilistic Robotics provides ways to deal with uncertainties when designing robotic systems.
- Sebastian Thrun, Wolfram Burgard, and Dieter Fox - Probabilistic Robotics: This book is a must read for state estimation and action selection in real robots, using probability techniques to overcome uncertainties in models, actuators, and sensors.
- Olivier Lebeltel - Programmation Bayesienne des Robots: Lebeltel presents the Bayesian Programming framework, a set of procedures to create architectures for robots that works in a probabilistic way from day zero. This work is in French only.
- Pierre Bessierre et. al. - Bayesian Programming: Working closely with Lebeltel, Bessiere presents Bayesian Programming for beginners in an easy -to-read book. The book also contains several examples to help understanding concepts in practice.
- Carla M. C. C. Koike - Bayesian Approach to Action Selection and Attention Focusing: Koike's work is closely related to the previous two, specially focusing on Bayesian Programming usage for action selection under uncertainties.
Reinforcement Learning is the learning approach used in this work for action selection (although Genetic Algorithms and State Machines are also used).
- Richard S. Sutton, Andrew G. Barto - Reinforcement Learning: This book is a no-brainer with respect to Reinforcement Learning studies. Written for people with no knowledge in the area, it is extremely simple to understand and apply Reinforcement Learning algorithms after reading this work.
- Chris Pryby, Charles Isbell, Michael Littman - Udacity's Reinforcement Learning Course If you prefer watching video classes over reading textbooks, this course is precious. Focused specifically in Reinforcement Learning, you should gain more than enough knowledge to work on any RL system after finishing it.
- Christopher J.C.H Watkins, Peter Dayan - Q-learning: This paper is a seminal work in Q-learning, one of the most famous Reinforcement Learning algorithms and used in this project.
- Marina Irodova, Robert H. Sloan - Reinforcement Learning and Function Approximation: Irodova and Sloan presents the challenges of using Q-learning with continuous environments and a method to adapt the basic algorithm for this use case.
Machine Learning is all about estimating from collected data in order to predict unseen cases. Many of its algorithms can be applied in several parts of a robot, such as in measurement interpretation, action selection, and state estimation and prediction.
- Emily Fox, Carlos Guestrin - Coursera's Machine Learning Specialization: This is the best possible way to start studying Machine Learning. Free to watch, this series of courses combines pretty well theory explanations and programming practices.
- Tom M. Mitchell - Machine Learning: Possibly one of the most famous books in Machine Learning, this book explains in great detail the origins of each Machine Learning technique and is definitely a great reference.
- Christopher Bishop - Pattern Recognition and Machine Learning: A undergraduate and early-graduate level introduction to Machine Learning, this is an interesting book to start studying Machine Learning.
Since many agents exist in the same environment, it configures a Multiagent System, which can be extremely different from usual, single agent systems. Understanding these differences are crucial to develop better techniques for our robots.
- Peter Stone, Manuela Veloso - Multiagent Systems: A Survey from a Machine Learning Perspective: This paper is a comprehensive explanation on multiagent systems and their implications to develop intelligent agents. It is strongly recommended to read this paper.
- Gerhard Weiss - Multiagent Systems - A Modern Approach to Distributed Artificial Intelligence: Weiss' book is an interesting first introduction to multiagent systems and challenges that arise when distributing Artificial Intelligence in several distinct computers or robots.
- Yoav Shoham, Kevin Leyton-Brown - Multiagent Systems: Algorithmic, Game-Theoretic, and Logical Foundations: Although more focused in the theory behind multiagent systems - reason why this book is used extensively in Game Theory and Economics classes - Shoham and Leyton-Brown's work answers fundamental questions when information is distributed through several agents in the same environment.