Reparameterization Policy Gradient (RPG) has emerged as a powerful paradigm for model-based reinforcement learning, enabling high sample efficiency by backpropagating gradients through differentiable dynamics. However, prior RPG approaches have been predominantly restricted to Gaussian policies, limiting their performance and failing to leverage recent advances in generative models. In this work, we identify that flow policies, which generate actions via differentiable ODE integration, naturally align with the RPG framework, a connection not established in prior work. However, naively exploiting this synergy proves ineffective, often suffering from training instability and a lack of exploration. We propose Reparameterization Flow Policy Optimization (RFO). RFO computes policy gradients by backpropagating jointly through the flow generation process and system dynamics, unlocking high sample efficiency without requiring intractable log-likelihood calculations. RFO includes two tailored regularization terms for stability and exploration. We also propose a variant of RFO with action chunking. Extensive experiments on diverse locomotion and manipulation tasks, involving both rigid and soft bodies with state or visual inputs, demonstrate the effectiveness of RFO. Notably, on a challenging locomotion task controlling a soft-body quadruped, RFO achieves almost $2\times$ the reward of the state-of-the-art baseline.
The current work on reinforcement learning (RL) from demonstrations often assumes the demonstrations are samples from an optimal policy, an unrealistic assumption in practice. When demonstrations are generated by sub-optimal policies or have sparse state-action pairs, policy learned from sub-optimal demonstrations may mislead an agent with incorrect or non-local action decisions. We propose a new method called Local Ensemble and Reparameterization with Split and Merge of expert policies (LEARN-SAM) to improve efficiency and make better use of the sub-optimal demonstrations. First, LEARN-SAM em…
We present weight normalization: a reparameterization of the weight vectors in a neural network that decouples the length of those weight vectors from their direction. By reparameterizing the weights in this way we improve the conditioning of the optimization problem and we speed up convergence of stochastic gradient descent. Our reparameterization is inspired by batch normalization but does not introduce any dependencies between the examples in a minibatch. This means that our method can also be applied successfully to recurrent models such as LSTMs and to noise-sensitive applications such as…
Reinforcement learning is well-studied under discrete actions. Integer actions setting is popular in the industry yet still challenging due to its high dimensionality. To this end, we study reinforcement learning under integer actions by incorporating the Soft Actor-Critic (SAC) algorithm with an integer reparameterization. Our key observation for integer actions is that their discrete structure can be simplified using their comparability property. Hence, the proposed integer reparameterization does not need one-hot encoding and is of low dimensionality. Experiments show that the proposed SAC …
Okinawa Institute of Science and Technology Graduate University（冲绳科学技术大学院大学）；University of Cambridge（剑桥大学）；TU Darmstadt（达姆施塔特工业大学）；Max Planck Institute for Intelligent Systems（马克斯·普朗克智能系统研究所）
Previously, the exploding gradient problem has been explained to be central in deep learning and model-based reinforcement learning, because it causes numerical issues and instability in optimization. Our experiments in model-based reinforcement learning imply that the problem is not just a numerical issue, but it may be caused by a fundamental chaos-like nature of long chains of nonlinear computations. Not only do the magnitudes of the gradients become large, the direction of the gradients becomes essentially random. We show that reparameterization gradients suffer from the problem, while lik…
Performative Prediction with Bandit Feedback: Learning through Reparameterization
University of California, Santa Cruz（加州大学圣克鲁兹分校）；Columbia University（哥伦比亚大学）；the Chinese University of Hong Kong（香港中文大学）；Washington University in St. Louis（圣路易斯华盛顿大学）
Distributional reinforcement learning (DRL) is a recent reinforcement learning framework whose success has been supported by various empirical studies. It relies on the key idea of replacing the expected return with the return distribution, which captures the intrinsic randomness of the long term rewards. Most of the existing literature on DRL focuses on problems with discrete action space and value based methods. In this work, motivated by applications in robotics with continuous action space control settings, we propose sample-based distributional policy gradient (SDPG) algorithm. It models …
State University of New York at Stony Brook（纽约州立大学石溪分校）；Peking University（北京大学）
Classical reinforcement learning (RL) aims to optimize the expected cumulative rewards. In this work, we consider the RL setting where the goal is to optimize the quantile of the cumulative rewards. We parameterize the policy controlling actions by neural networks and propose a novel policy gradient algorithm called Quantile-Based Policy Optimization (QPO) and its variant Quantile-Based Proximal Policy Optimization (QPPO) to solve deep RL problems with quantile objectives. QPO uses two coupled iterations running at different time scales for simultaneously estimating quantiles and policy parame…
Robotics Design Laboratory, University of Queensland（昆士兰大学机器人设计实验室）
Soft Actor-Critic (SAC) is one of the state-of-the-art off-policy reinforcement learning (RL) algorithms that is within the maximum entropy based RL framework. SAC is demonstrated to perform very well in a list of continous control tasks with good stability and robustness. SAC learns a stochastic Gaussian policy that can maximize a trade-off between total expected reward and the policy entropy. To update the policy, SAC minimizes the KL-Divergence between the current policy density and the soft value function density. Reparameterization trick is then used to obtain the approximate gradient of …
Shanghai Institute for Advanced Communication and Data Science（上海先进通信与数据科学研究院）；School of Computer Engineering and Science, Shanghai University（上海大学计算机工程与科学学院）
Instability and slowness are two main problems in deep reinforcement learning. Even if proximal policy optimization (PPO) is the state of the art, it still suffers from these two problems. We introduce an improved algorithm based on proximal policy optimization, mixed distributed proximal policy optimization (MDPPO), and show that it can accelerate and stabilize the training process. In our algorithm, multiple different policies train simultaneously and each of them controls several identical agents that interact with environments. Actions are sampled by each policy separately as usual, but th…
National Institute of Advanced Industrial Science and Technology（产业技术综合研究所）；NEC Central Research Laboratories（NEC中央研究所）；The University of Tokyo（东京大学）
Reinforcement Learning, a machine learning framework for training an autonomous agent based on rewards, has shown outstanding results in various domains. However, it is known that learning a good policy is difficult in a domain where rewards are rare. We propose a method, optimistic proximal policy optimization (OPPO) to alleviate this difficulty. OPPO considers the uncertainty of the estimated total return and optimistically evaluates the policy based on that amount. We show that OPPO outperforms the existing methods in a tabular task.
Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所）；School of Artificial Intelligence, University of Chinese Academy of Sciences（中国科学院大学人工智能学院）；University College London AI Lab（伦敦大学学院人工智能实验室）；Microsoft Research AI4Science（微软研究院AI4Science）；Li Auto（理想汽车）；The University of Edinburgh（爱丁堡大学）；Zhongguancun Institute of Artificial Intelligence（中关村人工智能研究院）
IOB: Integrating Optimization Transfer and Behavior Transfer for Multi-Policy Reuse
Faculty of Computing, Harbin Institute of Technology（哈尔滨工业大学计算机学院）；School of Artificial Intelligence, Optics and Electronics (iOPEN), Northwestern Polytechnical University（西北工业大学人工智能与光学电子学院（iOPEN））；McKelvey School of Engineering, Washington University in St. Louis（圣路易斯华盛顿大学麦凯维工程学院）
Reinforcement learning methods for continuous control tasks have evolved in recent years generating a family of policy gradient methods that rely primarily on a Gaussian distribution for modeling a stochastic policy. However, the Gaussian distribution has an infinite support, whereas real world applications usually have a bounded action space. This dissonance causes an estimation bias that can be eliminated if the Beta distribution is used for the policy instead, as it presents a finite support. In this work, we investigate how this Beta policy performs when it is trained by the Proximal Polic…
School of Computing, National University of Singapore, Singapore（新加坡国立大学计算机学院）；Department of Statistics, University of Oxford, United Kingdom（英国牛津大学统计系）；School of Intelligence Science（智能科学学院）；Technology, Nanjing University, China（技术学院，南京大学，中国）
As the most successful variant and improvement for Trust Region Policy Optimization (TRPO), proximal policy optimization (PPO) has been widely applied across various domains with several advantages: efficient data utilization, easy implementation, and good parallelism. In this paper, a first-order gradient reinforcement learning algorithm called Policy Optimization with Penalized Point Probability Distance (POP3D), which is a lower bound to the square of total variance divergence is proposed as another powerful variant. Firstly, we talk about the shortcomings of several commonly used algorithm…
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…
In real-world decision making tasks, it is critical for data-driven reinforcement learning methods to be both stable and sample efficient. On-policy methods typically generate reliable policy improvement throughout training, while off-policy methods make more efficient use of data through sample reuse. In this work, we combine the theoretically supported stability benefits of on-policy algorithms with the sample efficiency of off-policy algorithms. We develop policy improvement guarantees that are suitable for the off-policy setting, and connect these bounds to the clipping mechanism used in P…
Theoretical Analysis of Sparse Optimization with Reparameterization, Weight Decay, and Adaptive Learning Rate
State Key Laboratory of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China（人工智能安全国家重点实验室，计算技术研究所，中国科学院，北京，中国）；School of Computer Science and Technology, University of Chinese Academy of Sciences, Beijing, China（中国科学院大学计算机科学与技术学院，北京，中国）；Beijing Academy of Artificial Intelligence (BAAI), Beijing, China（北京人工智能研究院（BAAI），北京，中国）；IIIS, Tsinghua University, Beijing, China（清华大学人工智能院，北京，中国）；School of Statistics and Management, Shanghai University of Finance and Economics, Shanghai, China（上海财经大学统计与管理学院，上海，中国）；Institute of Data Science and Statistics, Shanghai University of Finance and Economics, Shanghai, China（上海财经大学数据科学与统计研究所，上海，中国）
Policy Optimization (PO) is a widely used approach to address continuous control tasks. In this paper, we introduce the notion of mediator feedback that frames PO as an online learning problem over the policy space. The additional available information, compared to the standard bandit feedback, allows reusing samples generated by one policy to estimate the performance of other policies. Based on this observation, we propose an algorithm, RANDomized-exploration policy Optimization via Multiple Importance Sampling with Truncation (RANDOMIST), for regret minimization in PO, that employs a randomi…
Technion Israel Institute of Technology（以色列理工学院）；DeepMind（深度思维）
Solving tasks in Reinforcement Learning is no easy feat. As the goal of the agent is to maximize the accumulated reward, it often learns to exploit loopholes and misspecifications in the reward signal resulting in unwanted behavior. While constraints may solve this issue, there is no closed form solution for general constraints. In this work we present a novel multi-timescale approach for constrained policy optimization, called `Reward Constrained Policy Optimization' (RCPO), which uses an alternative penalty signal to guide the policy towards a constraint satisfying one. We prove the converge…
Very recently proximal policy optimization (PPO) algorithms have been proposed as first-order optimization methods for effective reinforcement learning. While PPO is inspired by the same learning theory that justifies trust region policy optimization (TRPO), PPO substantially simplifies algorithm design and improves data efficiency by performing multiple epochs of \emph{clipped policy optimization} from sampled data. Although clipping in PPO stands for an important new mechanism for efficient and reliable policy update, it may fail to adaptively improve learning performance in accordance with …
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…
In order for reinforcement learning techniques to be useful in real-world decision making processes, they must be able to produce robust performance from limited data. Deep policy optimization methods have achieved impressive results on complex tasks, but their real-world adoption remains limited because they often require significant amounts of data to succeed. When combined with small sample sizes, these methods can result in unstable learning due to their reliance on high-dimensional sample-based estimates. In this work, we develop techniques to control the uncertainty introduced by these e…
Entropy regularization is an important idea in reinforcement learning, with great success in recent algorithms like Soft Q Network (SQN) and Soft Actor-Critic (SAC1). In this work, we extend this idea into the on-policy realm. We propose the soft policy gradient theorem (SPGT) for on-policy maximum entropy reinforcement learning. With SPGT, a series of new policy optimization algorithms are derived, such as SPG, SA2C, SA3C, SDDPG, STRPO, SPPO, SIMPALA and so on. We find that SDDPG is equivalent to SAC1. For policy gradient, the policy network is often represented as a Gaussian distribution wit…
New York University（纽约大学）；UC San Diego（加州大学圣地亚哥分校）；New York University Shanghai（上海纽约大学）
In reinforcement learning, an agent attempts to learn high-performing behaviors through interacting with the environment, such behaviors are often quantified in the form of a reward function. However some aspects of behavior-such as ones which are deemed unsafe and to be avoided-are best captured through constraints. We propose a novel approach called First Order Constrained Optimization in Policy Space (FOCOPS) which maximizes an agent's overall reward while ensuring the agent satisfies a set of cost constraints. Using data generated from the current policy, FOCOPS first finds the optimal upd…
Constrained Markov Decision Process (CMDP) is a natural framework for reinforcement learning tasks with safety constraints, where agents learn a policy that maximizes the long-term reward while satisfying the constraints on the long-term cost. A canonical approach for solving CMDPs is the primal-dual method which updates parameters in primal and dual spaces in turn. Existing methods for CMDPs only use on-policy data for dual updates, which results in sample inefficiency and slow convergence. In this paper, we propose a policy search method for CMDPs called Accelerated Primal-Dual Optimization …
Off-policy learning is a framework for evaluating and optimizing policies without deploying them, from data collected by another policy. Real-world environments are typically non-stationary and the offline learned policies should adapt to these changes. To address this challenge, we study the novel problem of off-policy optimization in piecewise-stationary contextual bandits. Our proposed solution has two phases. In the offline learning phase, we partition logged data into categorical latent states and learn a near-optimal sub-policy for each state. In the online deployment phase, we adaptivel…
Applied AI Institute, Deakin University（迪肯大学应用人工智能研究所）
We introduce a constrained optimization method for policy gradient reinforcement learning, which uses a virtual trust region to regulate each policy update. In addition to using the proximity of one single old policy as the normal trust region, we propose forming a second trust region through another virtual policy representing a wide range of past policies. We then enforce the new policy to stay closer to the virtual policy, which is beneficial if the old policy performs poorly. More importantly, we propose a mechanism to automatically build the virtual policy from a memory of past policies, …
Trust Region Policy Optimization (TRPO) is a popular and empirically successful policy search algorithm in reinforcement learning (RL). It iteratively solved the surrogate problem which restricts consecutive policies to be close to each other. TRPO is an on-policy algorithm. On-policy methods bring many benefits, like the ability to gauge each resulting policy. However, they typically discard all the knowledge about the policies which existed before. In this work, we use a replay buffer to borrow from the off-policy learning setting to TRPO. Entropy regularization is usually used to improve po…
Built on our previous study on green simulation assisted policy gradient (GS-PG) focusing on trajectory-based reuse, in this paper, we consider infinite-horizon Markov Decision Processes and create a new importance sampling based policy gradient optimization approach to support dynamic decision making. The existing GS-PG method was designed to learn from complete episodes or process trajectories, which limits its applicability to low-data situations and flexible online process control. To overcome this limitation, the proposed approach can selectively reuse the most related partial trajectorie…
DeepMind；California Institute of Technology（加州理工学院）；Mila；Université de Montréal（蒙特利尔大学）；University of Cambridge（剑桥大学）
Policy networks are a central feature of deep reinforcement learning (RL) algorithms for continuous control, enabling the estimation and sampling of high-value actions. From the variational inference perspective on RL, policy networks, when used with entropy or KL regularization, are a form of \textit{amortized optimization}, optimizing network parameters rather than the policy distributions directly. However, \textit{direct} amortized mappings can yield suboptimal policy estimates and restricted distributions, limiting performance and exploration. Given this perspective, we consider the more …
We present an analytical policy update rule that is independent of parametric function approximators. The policy update rule is suitable for optimizing general stochastic policies and has a monotonic improvement guarantee. It is derived from a closed-form solution to trust-region optimization using calculus of variation, following a new theoretical result that tightens existing bounds for policy improvement using trust-region methods. The update rule builds a connection between policy search methods and value function methods. Moreover, off-policy reinforcement learning algorithms can be deriv…
Max Planck Institute for Human Cognitive and Brain Sciences（马克斯·普朗克人类认知与脑科学研究所）；Center for Scalable Data Analytics and Artificial Intelligence (ScaDS.AI)（可扩展数据与人工智能中心）；Institut für Mathematik, Technische Universität Berlin（柏林技术大学数学研究所）
We propose a new family of policy gradient methods for reinforcement learning, which alternate between sampling data through interaction with the environment, and optimizing a "surrogate" objective function using stochastic gradient ascent. Whereas standard policy gradient methods perform one gradient update per data sample, we propose a novel objective function that enables multiple epochs of minibatch updates. The new methods, which we call proximal policy optimization (PPO), have some of the benefits of trust region policy optimization (TRPO), but they are much simpler to implement, more ge…
