多步截断优先扫描强化学习算法

研究了优先扫描的强化学习方法,通过定义新的迹,把多步截断即时差分学习用于集成规划的优先扫描强化学习,用多步截断即时差分来定义扫描优先权,提出一种改进的优先扫描强化学习算法并进行仿真实验,实验结果表明,新算法的学习效率有明显的提高。     

Risk-Sensitive Reinforcement Learning

Most reinforcement learning algorithms optimize the expected return of a Markov Decision Problem. Practice has taught us the lesson that this criterion is not always the most suitable because many applications require robust control strategies which also take into account the variance of the return. Classical control literature provides several techniques to deal with risk-sensitive optimization goals like the so-called worst-case optimality criterion exclusively focusing on risk-avoiding policies or classical risk-sensitive control, which transforms the returns by exponential utility functions. While the first approach is typically too restrictive, the latter suffers from the absence of an obvious way to design a corresponding model-free reinforcement learning algorithm.Our risk-sensitive reinforcement learning algorithm is based on a very different philosophy. Instead of transforming the return of the process, we transform the temporal differences during learning. While our approach reflects important properties of the classical exponential utility framework, we avoid its serious drawbacks for learning. Based on an extended set of optimality equations we are able to formulate risk-sensitive versions of various well-known reinforcement learning algorithms which converge with probability one under the usual conditions.     

一类基于启发式搜索的激励学习算法

激励学习已被证明是在控制领域中一种可行的新方法。相比其他的方法,它能较好地处理未知环境问题,但它仍然不是一种有效的方法。幸运的是,在现实世界中,智能体总是会有一些环境的先验知识,这些能形成启发式信息。启发式搜索是一种常用的搜索方法,有很快的搜索速度,但需要精确的启发式信息,这在有些时候难以得到。文中分析比较了启发式搜索和激励学习的各自特点,提出一类新的基于启发式搜索的激励学习算法,初步的实验结果显示了较好的性能。     

Discounted Reinforcement Learning Does Not Scale

Currently popular reinforcement learning methods are based on estimating value functions that indicate the long–term value of each problem state. In many domains, such as those traditionally studied in AI planning research, the size of state spaces precludes the individual storage of state value estimates. Consequently, most practical implementations of reinforcement learning methods have stored value functions using generalizing function approximators, with mixed results. We analyze the effects of approximation error on the performance of goal–based tasks, revealing potentially severe scaling difficulties. Empirical evidence is presented that suggests when difficulties are likely to occur and explains some of the widely differing results reported in the literature.     

Robot Control Optimization Using Reinforcement Learning

Conventional robot control schemes are basically model-based methods. However, exact modeling of robot dynamics poses considerable problems and faces various uncertainties in task execution. This paper proposes a reinforcement learning control approach for overcoming such drawbacks. An artificial neural network (ANN) serves as the learning structure, and an applied stochastic real-valued (SRV) unit as the learning method. Initially, force tracking control of a two-link robot arm is simulated to verify the control design. The simulation results confirm that even without information related to the robot dynamic model and environment states, operation rules for simultaneous controlling force and velocity are achievable by repetitive exploration. Hitherto, however, an acceptable performance has demanded many learning iterations and the learning speed proved too slow for practical applications. The approach herein, therefore, improves the tracking performance by combining a conventional controller with a reinforcement learning strategy. Experimental results demonstrate improved trajectory tracking performance of a two-link direct-drive robot manipulator using the proposed method.     

Reinforcement Learning with Replacing Eligibility Traces

The eligibility trace is one of the basic mechanisms used in reinforcement learning to handle delayed reward. In this paper we introduce a new kind of eligibility trace, the replacing trace, analyze it theoretically, and show that it results in faster, more reliable learning than the conventional trace. Both kinds of trace assign credit to prior events according to how recently they occurred, but only the conventional trace gives greater credit to repeated events. Our analysis is for conventional and replace-trace versions of the offline TD(1) algorithm applied to undiscounted absorbing Markov chains. First, we show that these methods converge under repeated presentations of the training set to the same predictions as two well known Monte Carlo methods. We then analyze the relative efficiency of the two Monte Carlo methods. We show that the method corresponding to conventional TD is biased, whereas the method corresponding to replace-trace TD is unbiased. In addition, we show that the method corresponding to replacing traces is closely related to the maximum likelihood solution for these tasks, and that its mean squared error is always lower in the long run. Computational results confirm these analyses and show that they are applicable more generally. In particular, we show that replacing traces significantly improve performance and reduce parameter sensitivity on the "Mountain-Car" task, a full reinforcement-learning problem with a continuous state space, when using a feature-based function approximator.     

一种基于递归最小二乘法的强化学习算法及其应用研究

文章推导了递归最小二乘瞬时差分法,较通常的瞬时差分法有样本使用效率高,收敛速度快,计算量少等特点。并将基于递归最小二乘的强化学习应用于船舶航向控制,克服了通常智能算法的学习需要一定数量样本数据的缺陷,对控制器的参数进行在线学习与调整,可以在一定程度上解决船舶运动中的不确定性问题,仿真结果表明,在有各种分浪流干扰的条件下,船舶航向的控制仍能取得令人满意的效果,说明该算法是有效可行的。     

Reinforcement Learning and Robust Control for Robot Compliance Tasks

The complexity in planning and control of robot compliance tasks mainly results from simultaneous control of both position and force and inevitable contact with environments. It is quite difficult to achieve accurate modeling of the interaction between the robot and the environment during contact. In addition, the interaction with the environment varies even for compliance tasks of the same kind. To deal with these phenomena, in this paper, we propose a reinforcement learning and robust control scheme for robot compliance tasks. A reinforcement learning mechanism is used to tackle variations among compliance tasks of the same kind. A robust compliance controller that guarantees system stability in the presence of modeling uncertainties and external disturbances is used to execute control commands sent from the reinforcement learning mechanism. Simulations based on deburring compliance tasks demonstrate the effectiveness of the proposed scheme.     

结合LSTM的强化学习动态环境路径规划算法

在路径规划领域已经涌现出了诸多的优秀的经典算法,但这些传统方法往往基于静态环境,对于动态可变环境缺乏处理能力.本文提出一种结合LSTM强化学习动态环境路径规划算法.首先,本文以环境图像作为输入,最大限度了保证了原始的信息来源.而后构建了自动编码器用来对环境图像进行特征降维,降低了整体模型的复杂程度.最后采用深度强化学习...     

Module-Based Reinforcement Learning: Experiments with a Real Robot

The behavior of reinforcement learning (RL) algorithms is best understood in completely observable, discrete-time controlled Markov chains with finite state and action spaces. In contrast, robot-learning domains are inherently continuous both in time and space, and moreover are partially observable. Here we suggest a systematic approach to solve such problems in which the available qualitative and quantitative knowledge is used to reduce the complexity of learning task. The steps of the design process are to: (i) decompose the task into subtasks using the qualitative knowledge at hand; (ii) design local controllers to solve the subtasks using the available quantitative knowledge, and (iii) learn a coordination of these controllers by means of reinforcement learning. It is argued that the approach enables fast, semi-automatic, but still high quality robot-control as no fine-tuning of the local controllers is needed. The approach was verified on a non-trivial real-life robot task. Several RL algorithms were compared by ANOVA and it was found that the model-based approach worked significantly better than the model-free approach. The learnt switching strategy performed comparably to a handcrafted version. Moreover, the learnt strategy seemed to exploit certain properties of the environment which were not foreseen in advance, thus supporting the view that adaptive algorithms are advantageous to nonadaptive ones in complex environments.     

Module-Based Reinforcement Learning: Experiments with a Real Robot

The behavior of reinforcement learning (RL) algorithms is best understood in completely observable, discrete-time controlled Markov chains with finite state and action spaces. In contrast, robot-learning domains are inherently continuous both in time and space, and moreover are partially observable. Here we suggest a systematic approach to solve such problems in which the available qualitative and quantitative knowledge is used to reduce the complexity of learning task. The steps of the design process are to:i) decompose the task into subtasks using the qualitative knowledge at hand; ii) design local controllers to solve the subtasks using the available quantitative knowledge and iii) learn a coordination of these controllers by means of reinforcement learning. It is argued that the approach enables fast, semi-automatic, but still high quality robot-control as no fine-tuning of the local controllers is needed. The approach was verified on a non-trivial real-life robot task. Several RL algorithms were compared by ANOVA and it was found that the model-based approach worked significantly better than the model-free approach. The learnt switching strategy performed comparably to a handcrafted version. Moreover, the learnt strategy seemed to exploit certain properties of the environment which were not foreseen in advance, thus supporting the view that adaptive algorithms are advantageous to non-adaptive ones in complex environments.     

Technical Note: Q-Learning

Q-learning (Watkins, 1989) is a simple way for agents to learn how to act optimally in controlled Markovian domains. It amounts to an incremental method for dynamic programming which imposes limited computational demands. It works by successively improving its evaluations of the quality of particular actions at particular states.This paper presents and proves in detail a convergence theorem forQ-learning based on that outlined in Watkins (1989). We show thatQ-learning converges to the optimum action-values with probability 1 so long as all actions are repeatedly sampled in all states and the action-values are represented discretely. We also sketch extensions to the cases of non-discounted, but absorbing, Markov environments, and where manyQ values can be changed each iteration, rather than just one.     

Associative Reinforcement Learning: A Generate and Test Algorithm

An agent that must learn to act in the world by trial and error faces thereinforcement learning problem, which is quite different from standard concept learning. Although good algorithms exist for this problem in the general case, they are often quite inefficient and do not exhibit generalization. One strategy is to find restricted classes of action policies that can be learned more efficiently. This paper pursues that strategy by developing an algorithm that performans an on-line search through the space of action mappings, expressed as Boolean formulae. The algorithm is compared with existing methods in empirical trials and is shown to have very good performance.     

A Study of Reinforcement Learning in the Continuous Case by the Means of Viscosity Solutions

This paper proposes a study of Reinforcement Learning (RL) for continuous state-space and time control problems, based on the theoretical framework of viscosity solutions (VSs). We use the method of dynamic programming (DP) which introduces the value function (VF), expectation of the best future cumulative reinforcement. In the continuous case, the value function satisfies a non-linear first (or second) order (depending on the deterministic or stochastic aspect of the process) differential equation called the Hamilton-Jacobi-Bellman (HJB) equation. It is well known that there exists an infinity of generalized solutions (differentiable almost everywhere) to this equation, other than the VF. We show that gradient-descent methods may converge to one of these generalized solutions, thus failing to find the optimal control.In order to solve the HJB equation, we use the powerful framework of viscosity solutions and state that there exists a unique viscosity solution to the HJB equation, which is the value function. Then, we use another main result of VSs (their stability when passing to the limit) to prove the convergence of numerical approximations schemes based on finite difference (FD) and finite element (FE) methods. These methods discretize, at some resolution, the HJB equation into a DP equation of a Markov Decision Process (MDP), which can be solved by DP methods (thanks to a strong contraction property) if all the initial data (the state dynamics and the reinforcement function) were perfectly known. However, in the RL approach, as we consider a system in interaction with some a priori (at least partially) unknown environment, which learns from experience, the initial data are not perfectly known but have to be approximated during learning. The main contribution of this work is to derive a general convergence theorem for RL algorithms when one uses only approximations (in a sense of satisfying some weak contraction property) of the initial data. This result can be used for model-based or model-free RL algorithms, with off-line or on-line updating methods, for deterministic or stochastic state dynamics (though this latter case is not described here), and based on FE or FD discretization methods. It is illustrated with several RL algorithms and one numerical simulation for the Car on the Hill problem.     

Learning Bayesian Networks: The Combination of Knowledge and Statistical Data

We describe a Bayesian approach for learning Bayesian networks from a combination of prior knowledge and statistical data. First and foremost, we develop a methodology for assessing informative priors needed for learning. Our approach is derived from a set of assumptions made previously as well as the assumption of likelihood equivalence, which says that data should not help to discriminate network structures that represent the same assertions of conditional independence. We show that likelihood equivalence when combined with previously made assumptions implies that the user's priors for network parameters can be encoded in a single Bayesian network for the next case to be seen—a prior network—and a single measure of confidence for that network. Second, using these priors, we show how to compute the relative posterior probabilities of network structures given data. Third, we describe search methods for identifying network structures with high posterior probabilities. We describe polynomial algorithms for finding the highest-scoring network structures in the special case where every node has at most k = 1 parent. For the general case (k > 1), which is NP-hard, we review heuristic search algorithms including local search, iterative local search, and simulated annealing. Finally, we describe a methodology for evaluating Bayesian-network learning algorithms, and apply this approach to a comparison of various approaches.     

采用核增强学习方法的多机器人编队控制

提出一种分布式的核增强学习方法来优化多机器人编队控制性能.首先,通过添加虚拟领队机器人,结合分布式的跟随控制策略,实现基本的多机器人编队控制:其次,提出结合最小二乘策略迭代和策略评测的核增强学习方法,即利用基于核的最小二乘策略迭代算法离线获取初始的编队优化控制策略,再利用基于核的最小二乘策略计测算法实现编队控制策略的在...     

Building a Basic Block Instruction Scheduler with Reinforcement Learning and Rollouts

The execution order of a block of computer instructions on a pipelined machine can make a difference in running time by a factor of two or more. Compilers use heuristic schedulers appropriate to each specific architecture implementation to achieve the best possible program speed. However, these heuristic schedulers are time-consuming and expensive to build. We present empirical results using both rollouts and reinforcement learning to construct heuristics for scheduling basic blocks. In simulation, the rollout scheduler outperformed a commercial scheduler on all benchmarks tested, and the reinforcement learning scheduler outperformed the commercial scheduler on several benchmarks and performed well on the others. The combined reinforcement learning and rollout approach was also very successful. We present results of running the schedules on Compaq Alpha machines and show that the results from the simulator correspond well to the actual run-time results.     

Distributed Deep Reinforcement Learning:A Survey and a Multi-player Multi-agent Learning Toolbox

With the breakthrough of AlphaGo,deep reinforcement learning has become a recognized technique for solving sequential decision-making problems.Despite its reputation,data inefficiency caused by its trial and error learning mechanism makes deep rein-forcement learning difficult to apply in a wide range of areas.Many methods have been developed for sample efficient deep reinforce-ment learning,such as environment modelling,experience transfer,and distributed modifications,among which distributed deep rein-forcement learning has shown its potential in various applications,such as human-computer gaming and intelligent transportation.In this paper,we conclude the state of this exciting field,by comparing the classical distributed deep reinforcement learning methods and studying important components to achieve efficient distributed learning,covering single player single agent distributed deep reinforce-ment learning to the most complex multiple players multiple agents distributed deep reinforcement learning.Furthermore,we review re-cently released toolboxes that help to realize distributed deep reinforcement learning without many modifications of their non-distrib-uted versions.By analysing their strengths and weaknesses,a multi-player multi-agent distributed deep reinforcement learning toolbox is developed and released,which is further validated on Wargame,a complex environment,showing the usability of the proposed tool-box for multiple players and multiple agents distributed deep reinforcement learning under complex games.Finally,we try to point out challenges and future trends,hoping that this brief review can provide a guide or a spark for researchers who are interested in distrib-uted deep reinforcement learning.     

Enhancing Time Series Clustering by Incorporating Multiple Distance Measures with Semi-Supervised Learning

Time series clustering is widely applied in various areas. Existing researches focus mainly on distance measures between two time series, such as dynamic time warping (DTW) based methods, edit-distance...