We consider the infinite-horizon discounted optimal control problem formalized by Markov Decision Processes. We focus on Policy Search algorithms, that compute an approximately optimal policy by following the standard Policy Iteration (PI) scheme via an -approximate greedy operator (Kakade and Langford, 2002; Lazaric et al., 2010). We describe existing and a few new performance bounds for Direct Policy Iteration (DPI) (Lagoudakis and Parr, 2003; Fern et al., 2006; Lazaric et al., 2010) and Conservative Policy Iteration (CPI) (Kakade and Langford, 2002). By paying a particular attention to the concentrability constants involved in such guarantees, we notably argue that the guarantee of CPI is much better than that of DPI, but this comes at the cost of a relative--exponential in $\frac{1}ε$-- increase of time complexity. We then describe an algorithm, Non-Stationary Direct Policy Iteration (NSDPI), that can either be seen as 1) a variation of Policy Search by Dynamic Programming by Bagnell et al. (2003) to the infinite horizon situation or 2) a simplified version of the Non-Stationary PI with growing period of Scherrer and Lesner (2012). We provide an analysis of this algorithm, that shows in particular that it enjoys the best of both worlds: its performance guarantee is similar to that of CPI, but within a time complexity similar to that of DPI.
Nara Institute of Science and Technology（奈良先端科学技术大学院大学）
In this paper, we propose cautious policy programming (CPP), a novel value-based reinforcement learning (RL) algorithm that can ensure monotonic policy improvement during learning. Based on the nature of entropy-regularized RL, we derive a new entropy regularization-aware lower bound of policy improvement that only requires estimating the expected policy advantage function. CPP leverages this lower bound as a criterion for adjusting the degree of a policy update for alleviating policy oscillation. Different from similar algorithms that are mostly theory-oriented, we also propose a novel interp…
Decisions in public health are almost always made in the context of uncertainty. Policy makers are responsible for making important decisions, faced with the daunting task of choosing from amongst many possible options. This task is called planning under uncertainty, and is particularly acute when addressing complex systems, such as issues of global health and development. Uncertainty leads to cautious or incorrect decisions that cost time, money, and human life. It is with this understanding that we pursue greater clarity on, and methods to address optimal policy making in health. Decision ma…
Okinawa Institute of Science and Technology（冲绳科学技术大学院大学）；ATR Computational Neuroscience Laboratories（ATR计算神经科学实验室）
Approximate dynamic programming algorithms, such as approximate value iteration, have been successfully applied to many complex reinforcement learning tasks, and a better approximate dynamic programming algorithm is expected to further extend the applicability of reinforcement learning to various tasks. In this paper we propose a new, robust dynamic programming algorithm that unifies value iteration, advantage learning, and dynamic policy programming. We call it generalized value iteration (GVI) and its approximated version, approximate GVI (AGVI). We show AGVI's performance guarantee, which i…
Nanyang Technological University（南洋理工大学）；Shandong University（山东大学）；Singapore Institute of Manufacturing Technology（新加坡制造技术研究院）
We propose a deep reinforcement learning (RL) method to learn large neighborhood search (LNS) policy for integer programming (IP). The RL policy is trained as the destroy operator to select a subset of variables at each step, which is reoptimized by an IP solver as the repair operator. However, the combinatorial number of variable subsets prevents direct application of typical RL algorithms. To tackle this challenge, we represent all subsets by factorizing them into binary decisions on each variable. We then design a neural network to learn policies for each variable in parallel, trained by a …
A common paradigm in classical planning is heuristic forward search. Forward search planners often rely on simple best-first search which remains fixed throughout the search process. In this paper, we introduce a novel search framework capable of alternating between several forward search approaches while solving a particular planning problem. Selection of the approach is performed using a trainable stochastic policy, mapping the state of the search to a probability distribution over the approaches. This enables using policy gradient to learn search strategies tailored to a specific distributi…
We consider infinite horizon dynamic programming problems, where the control at each stage consists of several distinct decisions, each one made by one of several agents. In an earlier work we introduced a policy iteration algorithm, where the policy improvement is done one-agent-at-a-time in a given order, with knowledge of the choices of the preceding agents in the order. As a result, the amount of computation for each policy improvement grows linearly with the number of agents, as opposed to exponentially for the standard all-agents-at-once method. For the case of a finite-state discounted …
Microsoft Research（微软研究院）；IBM Research AI（IBM人工智能研究院）；Singapore Management University（新加坡管理大学）
We propose a framework, called neural-progressive hedging (NP), that leverages stochastic programming during the online phase of executing a reinforcement learning (RL) policy. The goal is to ensure feasibility with respect to constraints and risk-based objectives such as conditional value-at-risk (CVaR) during the execution of the policy, using probabilistic models of the state transitions to guide policy adjustments. The framework is particularly amenable to the class of sequential resource allocation problems since feasibility with respect to typical resource constraints cannot be enforced …
This paper presents a constrained adaptive dynamic programming (CADP) algorithm to solve general nonlinear nonaffine optimal control problems with known dynamics. Unlike previous ADP algorithms, it can directly deal with problems with state constraints. Firstly, a constrained generalized policy iteration (CGPI) framework is developed to handle state constraints by transforming the traditional policy improvement process into a constrained policy optimization problem. Next, we propose an actor-critic variant of CGPI, called CADP, in which both policy and value functions are approximated by multi…
Shanghai Jiao Tong University（上海交通大学）；Huawei Noah’s Ark Lab（华为诺亚方舟实验室）；University College London（伦敦大学学院）
Deriving a good variable selection strategy in branch-and-bound is essential for the efficiency of modern mixed-integer programming (MIP) solvers. With MIP branching data collected during the previous solution process, learning to branch methods have recently become superior over heuristics. As branch-and-bound is naturally a sequential decision making task, one should learn to optimize the utility of the whole MIP solving process instead of being myopic on each step. In this work, we formulate learning to branch as an offline reinforcement learning (RL) problem, and propose a long-sighted hyb…
The combination of Monte-Carlo Tree Search (MCTS) and deep reinforcement learning is state-of-the-art in two-player perfect-information games. In this paper, we describe a search algorithm that uses a variant of MCTS which we enhanced by 1) a novel action value normalization mechanism for games with potentially unbounded rewards (which is the case in many optimization problems), 2) defining a virtual loss function that enables effective search parallelization, and 3) a policy network, trained by generations of self-play, to guide the search. We gauge the effectiveness of our method in "SameGam…
Korea Advanced Institute of Science and Technology（韩国高等科学技术研究院）
In this paper, a new population-guided parallel learning scheme is proposed to enhance the performance of off-policy reinforcement learning (RL). In the proposed scheme, multiple identical learners with their own value-functions and policies share a common experience replay buffer, and search a good policy in collaboration with the guidance of the best policy information. The key point is that the information of the best policy is fused in a soft manner by constructing an augmented loss function for policy update to enlarge the overall search region by the multiple learners. The guidance by th…
The University of Texas at Austin（德克萨斯大学奥斯汀分校）；The University of Massachusetts（马萨诸塞大学）；Carnegie Mellon University（卡内基梅隆大学）
We consider the task of evaluating a policy for a Markov decision process (MDP). The standard unbiased technique for evaluating a policy is to deploy the policy and observe its performance. We show that the data collected from deploying a different policy, commonly called the behavior policy, can be used to produce unbiased estimates with lower mean squared error than this standard technique. We derive an analytic expression for the optimal behavior policy --- the behavior policy that minimizes the mean squared error of the resulting estimates. Because this expression depends on terms that are…
Institut d’Investigació en Intelligència Artificial（人工智能研究所）；Universidade Metodista de São Paulo（圣保罗卫理公会大学）；Centro Universitário FEI（FEI大学中心）
Non-stationary domains, where unforeseen changes happen, present a challenge for agents to find an optimal policy for a sequential decision making problem. This work investigates a solution to this problem that combines Markov Decision Processes (MDP) and Reinforcement Learning (RL) with Answer Set Programming (ASP) in a method we call ASP(RL). In this method, Answer Set Programming is used to find the possible trajectories of an MDP, from where Reinforcement Learning is applied to learn the optimal policy of the problem. Results show that ASP(RL) is capable of efficiently finding the optimal …
Sorbonne Université（索邦大学）；Institut des Systèmes Intelligents et de Robotique（智能系统与机器人研究所）；École Normale Supérieure（巴黎高等师范学院）；Institut Jean Nicod（让·尼科研究所）
This paper focuses on a class of reinforcement learning problems where significant events are rare and limited to a single positive reward per episode. A typical example is that of an agent who has to choose a partner to cooperate with, while a large number of partners are simply not interested in cooperating, regardless of what the agent has to offer. We address this problem in a continuous state and action space with two different kinds of search methods: a gradient policy search method and a direct policy search method using an evolution strategy. We show that when significant events are ra…
In this paper, we present a new algorithm for parallel Monte Carlo tree search (MCTS). It is based on the pipeline pattern and allows flexible management of the control flow of the operations in parallel MCTS. The pipeline pattern provides for the first structured parallel programming approach to MCTS. Moreover, we propose a new lock-free tree data structure for parallel MCTS which removes synchronization overhead. The Pipeline Pattern for Parallel MCTS algorithm (called 3PMCTS), scales very well to higher numbers of cores when compared to the existing methods.
Traditional model-based reinforcement learning approaches learn a model of the environment dynamics without explicitly considering how it will be used by the agent. In the presence of misspecified model classes, this can lead to poor estimates, as some relevant available information is ignored. In this paper, we introduce a novel model-based policy search approach that exploits the knowledge of the current agent policy to learn an approximate transition model, focusing on the portions of the environment that are most relevant for policy improvement. We leverage a weighting scheme, derived from…
University of Illinois at Urbana-Champaign（伊利诺伊大学厄巴纳-尚佩恩分校）；Sandia National Laboratories（桑迪亚国家实验室）；c3.ai DTI
Function approximation is widely used in reinforcement learning to handle the computational difficulties associated with very large state spaces. However, function approximation introduces errors which may lead to instabilities when using approximate dynamic programming techniques to obtain the optimal policy. Therefore, techniques such as lookahead for policy improvement and m-step rollout for policy evaluation are used in practice to improve the performance of approximate dynamic programming with function approximation. We quantitatively characterize, for the first time, the impact of lookah…
University of Southern California（南加州大学）；Google Brain（谷歌大脑）
In principle, reinforcement learning and policy search methods can enable robots to learn highly complex and general skills that may allow them to function amid the complexity and diversity of the real world. However, training a policy that generalizes well across a wide range of real-world conditions requires far greater quantity and diversity of experience than is practical to collect with a single robot. Fortunately, it is possible for multiple robots to share their experience with one another, and thereby, learn a policy collectively. In this work, we explore distributed and asynchronous p…
California Institute of Technology（加州理工学院）；Jet Propulsion Laboratory, California Institute of Technology（加州理工学院喷气推进实验室）
We study the problem of learning sequential decision-making policies in settings with multiple state-action representations. Such settings naturally arise in many domains, such as planning (e.g., multiple integer programming formulations) and various combinatorial optimization problems (e.g., those with both integer programming and graph-based formulations). Inspired by the classical co-training framework for classification, we study the problem of co-training for policy learning. We present sufficient conditions under which learning from two views can improve upon learning from a single view …
Reinforcement learning policies are often represented by neural networks, but programmatic policies are preferred in some cases because they are more interpretable, amenable to formal verification, or generalize better. While efficient algorithms for learning neural policies exist, learning programmatic policies is challenging. Combining imitation-projection and dataset aggregation with a local search heuristic, we present a simple and direct approach to extracting a programmatic policy from a pretrained neural policy. After examining our local search heuristic on a programming by example prob…
Stanford University（斯坦福大学）；University of Illinois Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）；Nanjing University（南京大学）
Gradient-based methods are often used for policy optimization in deep reinforcement learning, despite being vulnerable to local optima and saddle points. Although gradient-free methods (e.g., genetic algorithms or evolution strategies) help mitigate these issues, poor initialization and local optima are still concerns in highly nonconvex spaces. This paper presents a method for policy optimization based on Monte-Carlo tree search and gradient-free optimization. Our method, called Monte-Carlo tree search for policy optimization (MCTSPO), provides a better exploration-exploitation trade-off thro…
Adapting an agent's behaviour to new environments has been one of the primary focus areas of physics based reinforcement learning. Although recent approaches such as universal policy networks partially address this issue by enabling the storage of multiple policies trained in simulation on a wide range of dynamic/latent factors, efficiently identifying the most appropriate policy for a given environment remains a challenge. In this work, we propose a Gaussian Process-based prior learned in simulation, that captures the likely performance of a policy when transferred to a previously unseen envi…
Local Policy Search is a popular reinforcement learning approach for handling large state spaces. Formally, it searches locally in a paramet erized policy space in order to maximize the associated value function averaged over some predefined distribution. It is probably commonly b elieved that the best one can hope in general from such an approach is to get a local optimum of this criterion. In this article, we show th e following surprising result: \emph{any} (approximate) \emph{local optimum} enjoys a \emph{global performance guarantee}. We compare this g uarantee with the one that is satisf…
We introduce a learning method called ``gradient-based reinforcement planning'' (GREP). Unlike traditional DP methods that improve their policy backwards in time, GREP is a gradient-based method that plans ahead and improves its policy before it actually acts in the environment. We derive formulas for the exact policy gradient that maximizes the expected future reward and confirm our ideas with numerical experiments.
DeepMind（深度思考）；OpenAI；University of California, Berkeley（加州大学伯克利分校）；Google Brain（谷歌大脑）
Monte Carlo Tree Search (MCTS) algorithms perform simulation-based search to improve policies online. During search, the simulation policy is adapted to explore the most promising lines of play. MCTS has been used by state-of-the-art programs for many problems, however a disadvantage to MCTS is that it estimates the values of states with Monte Carlo averages, stored in a search tree; this does not scale to games with very high branching factors. We propose an alternative simulation-based search method, Policy Gradient Search (PGS), which adapts a neural network simulation policy online via pol…
INRIA Bordeaux Sud-Ouest（法国国家信息与自动化研究所波尔多西南中心）；Sorbonne Université（索邦大学）；CNRS（法国国家科学研究中心）；German Aerospace Center (DLR)（德国航空航天中心）
Continuous action policy search is currently the focus of intensive research, driven both by the recent success of deep reinforcement learning algorithms and the emergence of competitors based on evolutionary algorithms. In this paper, we present a broad survey of policy search methods, providing a unified perspective on very different approaches, including also Bayesian Optimization and directed exploration methods. The main message of this overview is in the relationship between the families of methods, but we also outline some factors underlying sample efficiency properties of the various a…
It remains challenging to deploy existing risk-averse approaches to real-world applications. The reasons are multi-fold, including the lack of global optimality guarantee and the necessity of learning from long-term consecutive trajectories. Long-term consecutive trajectories are prone to involving visiting hazardous states, which is a major concern in the risk-averse setting. This paper proposes Short-Term VOlatility-controlled Policy Search (STOPS), a novel algorithm that solves risk-averse problems by learning from short-term trajectories instead of long-term trajectories. Short-term trajec…
University of Liège（列日大学）；INRIA Lille - Nord Europe（法国里尔-北欧信息与自动化研究所）
We propose novel policy search algorithms in the context of off-policy, batch mode reinforcement learning (RL) with continuous state and action spaces. Given a batch collection of trajectories, we perform off-line policy evaluation using an algorithm similar to that by [Fonteneau et al., 2010]. Using this Monte-Carlo like policy evaluator, we perform policy search in a class of parameterized policies. We propose both first order policy gradient and second order policy Newton algorithms. All our algorithms incorporate simultaneous perturbation estimates for the gradient as well as the Hessian o…
Deep Reinforcement Learning (DRL) is regarded as a potential method for car-following control and has been mostly studied to support a single following vehicle. However, it is more challenging to learn a stable and efficient car-following policy when there are multiple following vehicles in a platoon, especially with unpredictable leading vehicle behavior. In this context, we adopt an integrated DRL and Dynamic Programming (DP) approach to learn autonomous platoon control policies, which embeds the Deep Deterministic Policy Gradient (DDPG) algorithm into a finite-horizon value iteration framew…
We introduce two novel tree search algorithms that use a policy to guide search. The first algorithm is a best-first enumeration that uses a cost function that allows us to prove an upper bound on the number of nodes to be expanded before reaching a goal state. We show that this best-first algorithm is particularly well suited for `needle-in-a-haystack' problems. The second algorithm is based on sampling and we prove an upper bound on the expected number of nodes it expands before reaching a set of goal states. We show that this algorithm is better suited for problems where many paths lead to …
Guided policy search algorithms can be used to optimize complex nonlinear policies, such as deep neural networks, without directly computing policy gradients in the high-dimensional parameter space. Instead, these methods use supervised learning to train the policy to mimic a "teacher" algorithm, such as a trajectory optimizer or a trajectory-centric reinforcement learning method. Guided policy search methods provide asymptotic local convergence guarantees by construction, but it is not clear how much the policy improves within a small, finite number of iterations. We show that guided policy s…
Model-based Reinforcement Learning estimates the true environment through a world model in order to approximate the optimal policy. This family of algorithms usually benefits from better sample efficiency than their model-free counterparts. We investigate whether controllers learned in such a way are robust and able to generalize under small perturbations of the environment. Our work is inspired by the PILCO algorithm, a method for probabilistic policy search. We show that enforcing a lower bound to the likelihood noise in the Gaussian Process dynamics model regularizes the policy updates and …
MIT AI Laboratory（麻省理工学院人工智能实验室）；Brown University（布朗大学）
Cooperative games are those in which both agents share the same payoff structure. Value-based reinforcement-learning algorithms, such as variants of Q-learning, have been applied to learning cooperative games, but they only apply when the game state is completely observable to both agents. Policy search methods are a reasonable alternative to value-based methods for partially observable environments. In this paper, we provide a gradient-based distributed policy-search method for cooperative games and compare the notion of local optimum to that of Nash equilibrium. We demonstrate the effectiven…
Policy-gradient methods are widely used for learning control policies. They can be easily distributed to multiple workers and reach state-of-the-art results in many domains. Unfortunately, they exhibit large variance and subsequently suffer from high-sample complexity since they aggregate gradients over entire trajectories. At the other extreme, planning methods, like tree search, optimize the policy using single-step transitions that consider future lookahead. These approaches have been mainly considered for value-based algorithms. Planning-based algorithms require a forward model and are com…
We present a method for using adverb phrases to adjust skill parameters via learned adverb-skill groundings. These groundings allow an agent to use adverb feedback provided by a human to directly update a skill policy, in a manner similar to traditional local policy search methods. We show that our method can be used as a drop-in replacement for these policy search methods when dense reward from the environment is not available but human language feedback is. We demonstrate improved sample efficiency over modern policy search methods in two experiments.
We consider the problem of controlling a fully specified Markov decision process (MDP), also known as the planning problem, when the state space is very large and calculating the optimal policy is intractable. Instead, we pursue the more modest goal of optimizing over some small family of policies. Specifically, we show that the family of policies associated with a low-dimensional approximation of occupancy measures yields a tractable optimization. Moreover, we propose an efficient algorithm, scaling with the size of the subspace but not the state space, that is able to find a policy with low …
Local policy search is performed by most Deep Reinforcement Learning (D-RL) methods, which increases the risk of getting trapped in a local minimum. Furthermore, the availability of a simulation model is not fully exploited in D-RL even in simulation-based training, which potentially decreases efficiency. To better exploit simulation models in policy search, we propose to integrate a kinodynamic planner in the exploration strategy and to learn a control policy in an offline fashion from the generated environment interactions. We call the resulting model-based reinforcement learning method PPS …
Auburn University（奥本大学）；University of Iowa（爱荷华大学）；Facebook AI Research（Facebook 人工智能研究院）；Huawei Inc.（华为公司）
Off-policy policy optimization is a challenging problem in reinforcement learning (RL). The algorithms designed for this problem often suffer from high variance in their estimators, which results in poor sample efficiency, and have issues with convergence. A few variance-reduced on-policy policy gradient algorithms have been recently proposed that use methods from stochastic optimization to reduce the variance of the gradient estimate in the REINFORCE algorithm. However, these algorithms are not designed for the off-policy setting and are memory-inefficient, since they need to collect and stor…
Cooperative games are those in which both agents share the same payoff structure. Value-based reinforcement-learning algorithms, such as variants of Q-learning, have been applied to learning cooperative games, but they only apply when the game state is completely observable to both agents. Policy search methods are a reasonable alternative to value-based methods for partially observable environments. In this paper, we provide a gradient-based distributed policy-search method for cooperative games and compare the notion of local optimum to that of Nash equilibrium. We demonstrate the effectiven…
Learning policies on data synthesized by models can in principle quench the thirst of reinforcement learning algorithms for large amounts of real experience, which is often costly to acquire. However, simulating plausible experience de novo is a hard problem for many complex environments, often resulting in biases for model-based policy evaluation and search. Instead of de novo synthesis of data, here we assume logged, real experience and model alternative outcomes of this experience under counterfactual actions, actions that were not actually taken. Based on this, we propose the Counterfactua…
School of Computing, Informatics, and Decision Systems Engineering, Arizona State University（亚利桑那州立大学计算、信息与决策系统工程学院）
Despite the significant success at enabling robots with autonomous behaviors makes deep reinforcement learning a promising approach for robotic object search task, the deep reinforcement learning approach severely suffers from the nature sparse reward setting of the task. To tackle this challenge, we present a novel policy learning paradigm for the object search task, based on hierarchical and interpretable modeling with an intrinsic-extrinsic reward setting. More specifically, we explore the environment efficiently through a proxy low-level policy which is driven by the intrinsic rewarding su…
The combination of Monte-Carlo tree search (MCTS) with deep reinforcement learning has led to significant advances in artificial intelligence. However, AlphaZero, the current state-of-the-art MCTS algorithm, still relies on handcrafted heuristics that are only partially understood. In this paper, we show that AlphaZero's search heuristics, along with other common ones such as UCT, are an approximation to the solution of a specific regularized policy optimization problem. With this insight, we propose a variant of AlphaZero which uses the exact solution to this policy optimization problem, and …
Policy Search and Model Predictive Control~(MPC) are two different paradigms for robot control: policy search has the strength of automatically learning complex policies using experienced data, while MPC can offer optimal control performance using models and trajectory optimization. An open research question is how to leverage and combine the advantages of both approaches. In this work, we provide an answer by using policy search for automatically choosing high-level decision variables for MPC, which leads to a novel policy-search-for-model-predictive-control framework. Specifically, we formul…
We observe that several existing policy gradient methods (such as vanilla policy gradient, PPO, A2C) may suffer from overly large gradients when the current policy is close to deterministic (even in some very simple environments), leading to an unstable training process. To address this issue, we propose a new method, called \emph{target distribution learning} (TDL), for policy improvement in reinforcement learning. TDL alternates between proposing a target distribution and training the policy network to approach the target distribution. TDL is more effective in constraining the KL divergence …
In this paper we apply guided policy search (GPS) based reinforcement learning framework for a high dimensional optimal control problem arising in an additive manufacturing process. The problem comprises of controlling the process parameters so that layer-wise deposition of material leads to desired geometric characteristics of the resulting part surface while minimizing the material deposited. A realistic simulation model of the deposition process along with carefully selected set of guiding distributions generated based on iterative Linear Quadratic Regulator is used to train a neural networ…
University of Illinois at Urbana-Champaign（伊利诺伊大学厄巴纳-香槟分校）
Direct policy search has been widely applied in modern reinforcement learning and continuous control. However, the theoretical properties of direct policy search on nonsmooth robust control synthesis have not been fully understood. The optimal $\mathcal{H}_\infty$ control framework aims at designing a policy to minimize the closed-loop $\mathcal{H}_\infty$ norm, and is arguably the most fundamental robust control paradigm. In this work, we show that direct policy search is guaranteed to find the global solution of the robust $\mathcal{H}_\infty$ state-feedback control design problem. Notice th…
