Problem solving with reinforcement learning

This thesis is concerned with practical issues surrounding the application of reinforcement learning techniques to tasks that take place in high dimensional continuous state-space environments. In particular, the extension of on-line updating methods is considered, where the term implies systems that learn as each experience arrives, rather than storing the experiences for use in a separate oo-line learning phase. Firstly, the use of alternative update rules in place of standard Q-learning (Watkins 1989) is examined to provide faster convergence rates. Secondly, the use of multi-layer perceptron (MLP) neural networks (Rumelhart, Hinton and Williams 1986) is investigated to provide suitable generalising function approximators. Finally, consideration is given to the combination of Adaptive Heuristic Critic (AHC) methods and Q-learning to produce systems combining the beneets of real-valued actions and discrete switching. The diierent update rules examined are based on Q-learning combined with the TD() algorithm (Sutton 1988). Several new algorithms, including Modiied Q-Learning and Summation Q-Learning, are examined, as well as alternatives such as Q() (Peng and Williams 1994). In addition, algorithms are presented for applying these Q-learning updates to train MLPs on-line during trials, as opposed to the backward-replay method used by Lin (1993b) that requires waiting until the end of each trial before updating can occur. The performance of the update rules is compared on the Race Track problem of Barto, Bradtke and Singh (1993) using a lookup table representation for the Q-function. Some of the methods are found to perform almost as well as Real-Time Dynamic Programming, despite the fact that the latter has the advantage of a full world model. The performance of the connectionist algorithms is compared on a larger and more complex robot navigation problem. Here a simulated mobile robot is trained to guide itself to a goal position in the presence of obstacles. The robot must rely on limited sensory feedback from its surroundings and make decisions that can be generalised to arbitrary layouts of obstacles. These simulations show that the performance of on-line learning algorithms is less sensitive to the choice of training parameters than backward-replay, and that the alternative Q-learning rules of Modiied Q-Learning and Q() are more robust than standard Q-learning updates. Finally, a combination of real-valued AHC and Q-learning, called Q-AHC learning, is presented, and various architectures are compared in performance on the robot problem. The resulting reinforcement learning system has the properties of providing on-line training, parallel computation, generalising function …

[1]  M. Puterman,et al.  Modified Policy Iteration Algorithms for Discounted Markov Decision Problems , 1978 .

[2]  Richard S. Sutton,et al.  Neuronlike adaptive elements that can solve difficult learning control problems , 1983, IEEE Transactions on Systems, Man, and Cybernetics.

[3]  Richard S. Sutton,et al.  Temporal credit assignment in reinforcement learning , 1984 .

[4]  Rodney A. Brooks,et al.  A Robust Layered Control Syste For A Mobile Robot , 2022 .

[5]  S. Zucker,et al.  Toward Efficient Trajectory Planning: The Path-Velocity Decomposition , 1986 .

[6]  Dimitri P. Bertsekas,et al.  Dynamic Programming: Deterministic and Stochastic Models , 1987 .

[7]  Marcel Schoppers,et al.  Universal Plans for Reactive Robots in Unpredictable Environments , 1987, IJCAI.

[8]  Ken-ichi Funahashi,et al.  On the approximate realization of continuous mappings by neural networks , 1989, Neural Networks.

[9]  Kumpati S. Narendra,et al.  Learning automata - an introduction , 1989 .

[10]  John N. Tsitsiklis,et al.  Parallel and distributed computation , 1989 .

[11]  George Cybenko,et al.  Approximation by superpositions of a sigmoidal function , 1989, Math. Control. Signals Syst..

[12]  Michael I. Jordan,et al.  Learning to Control an Unstable System with Forward Modeling , 1989, NIPS.

[13]  Ronald J. Williams,et al.  A Learning Algorithm for Continually Running Fully Recurrent Neural Networks , 1989, Neural Computation.

[14]  Kurt Hornik,et al.  Multilayer feedforward networks are universal approximators , 1989, Neural Networks.

[15]  Paul J. Werbos,et al.  Backpropagation Through Time: What It Does and How to Do It , 1990, Proc. IEEE.

[16]  Oussama Khatib,et al.  Real-Time Obstacle Avoidance for Manipulators and Mobile Robots , 1985, Autonomous Robot Vehicles.

[17]  Richard S. Sutton,et al.  Integrated Architectures for Learning, Planning, and Reacting Based on Approximating Dynamic Programming , 1990, ML.

[18]  John C. Platt A Resource-Allocating Network for Function Interpolation , 1991, Neural Computation.

[19]  John E. W. Mayhew,et al.  Obstacle Avoidance through Reinforcement Learning , 1991, NIPS.

[20]  Michael I. Jordan,et al.  Hierarchies of Adaptive Experts , 1991, NIPS.

[21]  Carlos D. Brody,et al.  Fast Learning with Predictive Forward Models , 1991, NIPS.

[22]  Myung Won Kim,et al.  An efficient hidden node reduction technique for multilayer perceptrons , 1991, [Proceedings] 1991 IEEE International Joint Conference on Neural Networks.

[23]  Jean-Claude Latombe,et al.  Robot Motion Planning: A Distributed Representation Approach , 1991, Int. J. Robotics Res..

[24]  Sebastian Thrun,et al.  Active Exploration in Dynamic Environments , 1991, NIPS.

[25]  Qiuming Zhu,et al.  Hidden Markov model for dynamic obstacle avoidance of mobile robot navigation , 1991, IEEE Trans. Robotics Autom..

[26]  Michael I. Jordan,et al.  Task Decomposition Through Competition in a Modular Connectionist Architecture: The What and Where Vision Tasks , 1990, Cogn. Sci..

[27]  S. Thrun Eecient Exploration in Reinforcement Learning , 1992 .

[28]  Long-Ji Lin,et al.  Reinforcement learning for robots using neural networks , 1992 .

[29]  J. Millán,et al.  A Reinforcement Connectionist Approach to Robot Path Finding in Non-Maze-Like Environments , 2004, Machine Learning.

[30]  Steven J. Bradtke,et al.  Reinforcement Learning Applied to Linear Quadratic Regulation , 1992, NIPS.

[31]  Charles W. Anderson,et al.  Q-Learning with Hidden-Unit Restarting , 1992, NIPS.

[32]  Juan Carlos,et al.  Multistrategy Learning in a Reactive Control Systems for Autonomous Robotic Navigation , 1993, Informatica.

[33]  L.-J. Lin,et al.  Hierarchical learning of robot skills by reinforcement , 1993, IEEE International Conference on Neural Networks.

[34]  Russell Reed,et al.  Pruning algorithms-a survey , 1993, IEEE Trans. Neural Networks.

[35]  Leslie Pack Kaelbling,et al.  Learning in embedded systems , 1993 .

[36]  Martin Fodslette Møller,et al.  A scaled conjugate gradient algorithm for fast supervised learning , 1993, Neural Networks.

[37]  Wj Fitzgerald,et al.  Optimization schemes for neural networks , 1993 .

[38]  Ronald J. Williams,et al.  Tight Performance Bounds on Greedy Policies Based on Imperfect Value Functions , 1993 .

[39]  Anton Schwartz,et al.  A Reinforcement Learning Method for Maximizing Undiscounted Rewards , 1993, ICML.

[40]  Ronald J. Williams,et al.  Analysis of Some Incremental Variants of Policy Iteration: First Steps Toward Understanding Actor-Cr , 1993 .

[41]  Jing Peng,et al.  Efficient Learning and Planning Within the Dyna Framework , 1993, Adapt. Behav..

[42]  Long Ji Lin,et al.  Scaling Up Reinforcement Learning for Robot Control , 1993, International Conference on Machine Learning.

[43]  Gregory J. Wolff,et al.  Optimal Brain Surgeon and general network pruning , 1993, IEEE International Conference on Neural Networks.

[44]  Tony J. Prescott,et al.  Explorations in Reinforcement and Model-based Learning , 1994 .

[45]  LearningRichard S. Suttonsutton On Step-Size and Bias in Temporal-Di erence , 1994 .

[46]  Michael I. Jordan,et al.  MASSACHUSETTS INSTITUTE OF TECHNOLOGY ARTIFICIAL INTELLIGENCE LABORATORY and CENTER FOR BIOLOGICAL AND COMPUTATIONAL LEARNING DEPARTMENT OF BRAIN AND COGNITIVE SCIENCES , 1996 .

[47]  Sridhar Mahadevan,et al.  To Discount or Not to Discount in Reinforcement Learning: A Case Study Comparing R Learning and Q Learning , 1994, ICML.

[48]  Chen-Khong Tham,et al.  Modular on-line function approximation for scaling up reinforcement learning , 1994 .

[49]  Mahesan Niranjan,et al.  On-line Q-learning using connectionist systems , 1994 .

[50]  B Ravindran,et al.  A tutorial survey of reinforcement learning , 1994 .

[51]  Stewart W. Wilson ZCS: A Zeroth Level Classifier System , 1994, Evolutionary Computation.

[52]  V. Gullapalli,et al.  Acquiring robot skills via reinforcement learning , 1994, IEEE Control Systems.

[53]  Martin A. Riedmiller,et al.  Advanced supervised learning in multi-layer perceptrons — From backpropagation to adaptive learning algorithms , 1994 .

[54]  Ben J. A. Kröse,et al.  Learning from delayed rewards , 1995, Robotics Auton. Syst..

[55]  Pawea Cichosz Truncating Temporal Diierences: on the Eecient Implementation of Td() for Reinforcement Learning , 1995 .

[56]  Andrew G. Barto,et al.  Learning to Act Using Real-Time Dynamic Programming , 1995, Artif. Intell..

[57]  Richard S. Sutton,et al.  Reinforcement Learning with Replacing Eligibility Traces , 2005, Machine Learning.

[58]  Leslie Pack Kaelbling,et al.  On reinforcement learning for robots , 1996, IROS.

[59]  Thomas G. Dietterich What is machine learning? , 2020, Archives of Disease in Childhood.