One of the final frontiers in the development of complex human - AI collaborative systems is the ability of AI agents to comprehend the natural language and perform tasks accordingly. However, training efficient Reinforcement Learning (RL) agents grounded in natural language has been a long-standing challenge due to the complexity and ambiguity of the language and sparsity of the rewards, among other factors. Several advances in reinforcement learning, curriculum learning, continual learning, language models have independently contributed to effective training of grounded agents in various environments. Leveraging these developments, we present a novel algorithm, Grounded Language Instruction through DEmonstration in RL (GLIDE-RL) that introduces a teacher-instructor-student curriculum learning framework for training an RL agent capable of following natural language instructions that can generalize to previously unseen language instructions. In this multi-agent framework, the teacher and the student agents learn simultaneously based on the student's current skill level. We further demonstrate the necessity for training the student agent with not just one, but multiple teacher agents. Experiments on a complex sparse reward environment validates the effectiveness of our proposed approach.
With the continued introduction of driverless events to Formula:Society of Automotive Engineers (F:SAE) competitions around the world, teams are investigating all aspects of the autonomous vehicle stack. This paper presents the use of Deep Reinforcement Learning (DRL) and Inverse Reinforcement Learning (IRL) to map locally-observed cone positions to a desired steering angle for race track following. Two state-of-the-art algorithms not previously tested in this context: soft actor critic (SAC) and adversarial inverse reinforcement learning (AIRL), are used to train models in a representative simulation. Three novel reward functions for use by RL algorithms in an autonomous racing context are also discussed. Tests performed in simulation and the real world suggest that both algorithms can successfully train models for local path following. Suggestions for future work are presented to allow these models to scale to a full F:SAE vehicle.
With the increasing number of fast-electric vehicle charging stations (fast-EVCSs) and the popularization of information technology, electricity price competition between fast-EVCSs is highly expected, in which the utilization of public and/or privacy-preserved information will play a crucial role. Self-interest electric vehicle (EV) users, on the other hand, try to select a fast-EVCS for charging in a way to maximize their utilities based on electricity price, estimated waiting time, and their state of charge. While existing studies have largely focused on finding equilibrium prices, this study proposes a personalized dynamic pricing policy (PeDP) for a fast-EVCS to maximize revenue using a reinforcement learning (RL) approach. We first propose a multiple fast-EVCSs competing simulation environment to model the selfish behavior of EV users using a game-based charging station selection model with a monetary utility function. In the environment, we propose a Q-learning-based PeDP to maximize fast-EVCS' revenue. Through numerical simulations based on the environment: (1) we identify the importance of waiting time in the EV charging market by comparing the classic Bertrand competition model with the proposed PeDP for fast-EVCSs (from the system perspective); (2) we evaluate the performance of the proposed PeDP and analyze the effects of the information on the policy (from the service provider perspective); and (3) it can be seen that privacy-preserved information sharing can be misused by artificial intelligence-based PeDP in a certain situation in the EV charging market (from the customer perspective).
Reinforcement learning research obtained significant success and attention with the utilization of deep neural networks to solve problems in high dimensional state or action spaces. While deep reinforcement learning policies are currently being deployed in many different fields from medical applications to large language models, there are still ongoing questions the field is trying to answer on the generalization capabilities of deep reinforcement learning policies. In this paper, we will formalize and analyze generalization in deep reinforcement learning. We will explain the fundamental reasons why deep reinforcement learning policies encounter overfitting problems that limit their generalization capabilities. Furthermore, we will categorize and explain the manifold solution approaches to increase generalization, and overcome overfitting in deep reinforcement learning policies. From exploration to adversarial analysis and from regularization to robustness our paper provides an analysis on a wide range of subfields within deep reinforcement learning with a broad scope and in-depth view. We believe our study can provide a compact guideline for the current advancements in deep reinforcement learning, and help to construct robust deep neural policies with higher generalization skills.
In this paper, we aim to utilize only offline trajectory data to train a policy for multi-objective RL. We extend the offline policy-regularized method, a widely-adopted approach for single-objective offline RL problems, into the multi-objective setting in order to achieve the above goal. However, such methods face a new challenge in offline MORL settings, namely the preference-inconsistent demonstration problem. We propose two solutions to this problem: 1) filtering out preference-inconsistent demonstrations via approximating behavior preferences, and 2) adopting regularization techniques with high policy expressiveness. Moreover, we integrate the preference-conditioned scalarized update method into policy-regularized offline RL, in order to simultaneously learn a set of policies using a single policy network, thus reducing the computational cost induced by the training of a large number of individual policies for various preferences. Finally, we introduce Regularization Weight Adaptation to dynamically determine appropriate regularization weights for arbitrary target preferences during deployment. Empirical results on various multi-objective datasets demonstrate the capability of our approach in solving offline MORL problems.
Symbolic Music Alignment is the process of matching performed MIDI notes to corresponding score notes. In this paper, we introduce a reinforcement learning (RL)-based online symbolic music alignment technique. The RL agent - an attention-based neural network - iteratively estimates the current score position from local score and performance contexts. For this symbolic alignment task, environment states can be sampled exhaustively and the reward is dense, rendering a formulation as a simplified offline RL problem straightforward. We evaluate the trained agent in three ways. First, in its capacity to identify correct score positions for sampled test contexts; second, as the core technique of a complete algorithm for symbolic online note-wise alignment; and finally, as a real-time symbolic score follower. We further investigate the pitch-based score and performance representations used as the agent's inputs. To this end, we develop a second model, a two-step Dynamic Time Warping (DTW)-based offline alignment algorithm leveraging the same input representation. The proposed model outperforms a state-of-the-art reference model of offline symbolic music alignment.
Preference-based reinforcement learning (PBRL) in the offline setting has succeeded greatly in industrial applications such as chatbots. A two-step learning framework where one applies a reinforcement learning step after a reward modeling step has been widely adopted for the problem. However, such a method faces challenges from the risk of reward hacking and the complexity of reinforcement learning. To overcome the challenge, our insight is that both challenges come from the state-actions not supported in the dataset. Such state-actions are unreliable and increase the complexity of the reinforcement learning problem at the second step. Based on the insight, we develop a novel two-step learning method called PRC: preference-based reinforcement learning with constrained actions. The high-level idea is to limit the reinforcement learning agent to optimize over a constrained action space that excludes the out-of-distribution state-actions. We empirically verify that our method has high learning efficiency on various datasets in robotic control environments.
Multi-agent systems often require agents to collaborate with or compete against other agents with diverse goals, behaviors, or strategies. Agent modeling is essential when designing adaptive policies for intelligent machine agents in multiagent systems, as this is the means by which the ego agent understands other agents' behavior and extracts their meaningful policy representations. These representations can be used to enhance the ego agent's adaptive policy which is trained by reinforcement learning. However, existing agent modeling approaches typically assume the availability of local observations from other agents (modeled agents) during training or a long observation trajectory for policy adaption. To remove these constrictive assumptions and improve agent modeling performance, we devised a Contrastive Learning-based Agent Modeling (CLAM) method that relies only on the local observations from the ego agent during training and execution. With these observations, CLAM is capable of generating consistent high-quality policy representations in real-time right from the beginning of each episode. We evaluated the efficacy of our approach in both cooperative and competitive multi-agent environments. Our experiments demonstrate that our approach achieves state-of-the-art on both cooperative and competitive tasks, highlighting the potential of contrastive learning-based agent modeling for enhancing reinforcement learning.
Reinforcement learning (RL) is a powerful technique for training intelligent agents, but understanding why these agents make specific decisions can be quite challenging. This lack of transparency in RL models has been a long-standing problem, making it difficult for users to grasp the reasons behind an agent's behaviour. Various approaches have been explored to address this problem, with one promising avenue being reward decomposition (RD). RD is appealing as it sidesteps some of the concerns associated with other methods that attempt to rationalize an agent's behaviour in a post-hoc manner. RD works by exposing various facets of the rewards that contribute to the agent's objectives during training. However, RD alone has limitations as it primarily offers insights based on sub-rewards and does not delve into the intricate cause-and-effect relationships that occur within an RL agent's neural model. In this paper, we present an extension of RD that goes beyond sub-rewards to provide more informative explanations. Our approach is centred on a causal learning framework that leverages information-theoretic measures for explanation objectives that encourage three crucial properties of causal factors: causal sufficiency, sparseness, and orthogonality. These properties help us distill the cause-and-effect relationships between the agent's states and actions or rewards, allowing for a deeper understanding of its decision-making processes. Our framework is designed to generate local explanations and can be applied to a wide range of RL tasks with multiple reward channels. Through a series of experiments, we demonstrate that our approach offers more meaningful and insightful explanations for the agent's action selections.
Data assimilation (DA) plays a pivotal role in diverse applications, ranging from climate predictions and weather forecasts to trajectory planning for autonomous vehicles. A prime example is the widely used ensemble Kalman filter (EnKF), which relies on linear updates to minimize variance among the ensemble of forecast states. Recent advancements have seen the emergence of deep learning approaches in this domain, primarily within a supervised learning framework. However, the adaptability of such models to untrained scenarios remains a challenge. In this study, we introduce a novel DA strategy that utilizes reinforcement learning (RL) to apply state corrections using full or partial observations of the state variables. Our investigation focuses on demonstrating this approach to the chaotic Lorenz '63 system, where the agent's objective is to minimize the root-mean-squared error between the observations and corresponding forecast states. Consequently, the agent develops a correction strategy, enhancing model forecasts based on available system state observations. Our strategy employs a stochastic action policy, enabling a Monte Carlo-based DA framework that relies on randomly sampling the policy to generate an ensemble of assimilated realizations. Results demonstrate that the developed RL algorithm performs favorably when compared to the EnKF. Additionally, we illustrate the agent's capability to assimilate non-Gaussian data, addressing a significant limitation of the EnKF.
We present OpenRL, an advanced reinforcement learning (RL) framework designed to accommodate a diverse array of tasks, from single-agent challenges to complex multi-agent systems. OpenRL's robust support for self-play training empowers agents to develop advanced strategies in competitive settings. Notably, OpenRL integrates Natural Language Processing (NLP) with RL, enabling researchers to address a combination of RL training and language-centric tasks effectively. Leveraging PyTorch's robust capabilities, OpenRL exemplifies modularity and a user-centric approach. It offers a universal interface that simplifies the user experience for beginners while maintaining the flexibility experts require for innovation and algorithm development. This equilibrium enhances the framework's practicality, adaptability, and scalability, establishing a new standard in RL research. To delve into OpenRL's features, we invite researchers and enthusiasts to explore our GitHub repository at https://github.com/OpenRL-Lab/openrl and access our comprehensive documentation at https://openrl-docs.readthedocs.io.
This paper proposes a novel approach by integrating sensor fusion with deep reinforcement learning, specifically the Soft Actor-Critic (SAC) algorithm, to develop an optimal control policy for self-driving cars. Our system employs a two-branch fusion method for vehicle image and tracking sensor data, leveraging the strengths of residual structures and identity mapping to enhance agent training. Through comprehensive comparisons, we demonstrate the efficacy of information fusion and establish the superiority of our selected algorithm over alternative approaches. Our work advances the field of autonomous driving and demonstrates the potential of reinforcement learning in enabling intelligent vehicle decision-making.
Deep reinforcement learning (DRL) faces significant challenges in addressing the hard-exploration problems in tasks with sparse or deceptive rewards and large state spaces. These challenges severely limit the practical application of DRL. Most previous exploration methods relied on complex architectures to estimate state novelty or introduced sensitive hyperparameters, resulting in instability. To mitigate these issues, we propose an efficient adaptive trajectory-constrained exploration strategy for DRL. The proposed method guides the policy of the agent away from suboptimal solutions by leveraging incomplete offline demonstrations as references. This approach gradually expands the exploration scope of the agent and strives for optimality in a constrained optimization manner. Additionally, we introduce a novel policy-gradient-based optimization algorithm that utilizes adaptively clipped trajectory-distance rewards for both single- and multi-agent reinforcement learning. We provide a theoretical analysis of our method, including a deduction of the worst-case approximation error bounds, highlighting the validity of our approach for enhancing exploration. To evaluate the effectiveness of the proposed method, we conducted experiments on two large 2D grid world mazes and several MuJoCo tasks. The extensive experimental results demonstrate the significant advantages of our method in achieving temporally extended exploration and avoiding myopic and suboptimal behaviors in both single- and multi-agent settings. Notably, the specific metrics and quantifiable results further support these findings. The code used in the study is available at \url{https://github.com/buaawgj/TACE}.
We study a class of constrained reinforcement learning (RL) problems in which multiple constraint specifications are not identified before training. It is challenging to identify appropriate constraint specifications due to the undefined trade-off between the reward maximization objective and the constraint satisfaction, which is ubiquitous in constrained decision-making. To tackle this issue, we propose a new constrained RL approach that searches for policy and constraint specifications together. This method features the adaptation of relaxing the constraint according to a relaxation cost introduced in the learning objective. Since this feature mimics how ecological systems adapt to disruptions by altering operation, our approach is termed as resilient constrained RL. Specifically, we provide a set of sufficient conditions that balance the constraint satisfaction and the reward maximization in notion of resilient equilibrium, propose a tractable formulation of resilient constrained policy optimization that takes this equilibrium as an optimal solution, and advocate two resilient constrained policy search algorithms with non-asymptotic convergence guarantees on the optimality gap and constraint satisfaction. Furthermore, we demonstrate the merits and the effectiveness of our approach in computational experiments.
Learning policies that can generalize to unseen environments is a fundamental challenge in visual reinforcement learning (RL). While most current methods focus on acquiring robust visual representations through auxiliary supervision, pre-training, or data augmentation, the potential of modern vision foundation models remains underleveraged. In this work, we introduce Segment Anything Model for Generalizable visual RL (SAM-G), a novel framework that leverages the promptable segmentation ability of Segment Anything Model (SAM) to enhance the generalization capabilities of visual RL agents. We utilize image features from DINOv2 and SAM to find correspondence as point prompts to SAM, and then SAM produces high-quality masked images for agents directly. Evaluated across 8 DMControl tasks and 3 Adroit tasks, SAM-G significantly improves the visual generalization ability without altering the RL agents' architecture but merely their observations. Notably, SAM-G achieves 44% and 29% relative improvements on the challenging video hard setting on DMControl and Adroit respectively, compared to state-of-the-art methods. Video and code: https://yanjieze.com/SAM-G/
These lecture notes give a statistical perspective on the foundations of reinforcement learning and interactive decision making. We present a unifying framework for addressing the exploration-exploitation dilemma using frequentist and Bayesian approaches, with connections and parallels between supervised learning/estimation and decision making as an overarching theme. Special attention is paid to function approximation and flexible model classes such as neural networks. Topics covered include multi-armed and contextual bandits, structured bandits, and reinforcement learning with high-dimensional feedback.
Diffusion models have shown impressive performance in many domains. However, the model's capability to follow natural language instructions (e.g., spatial relationships between objects, generating complex scenes) is still unsatisfactory. In this work, we propose Iterative Prompt Relabeling (IPR), a novel algorithm that aligns images to text through iterative image sampling and prompt relabeling with feedback. IPR first samples a batch of images conditioned on the text, then relabels the text prompts of unmatched text-image pairs with classifier feedback. We conduct thorough experiments on SDv2 and SDXL, testing their capability to follow instructions on spatial relations. With IPR, we improved up to 15.22% (absolute improvement) on the challenging spatial relation VISOR benchmark, demonstrating superior performance compared to previous RL methods. Our code is publicly available at https://github.com/cxy000000/IPR-RLDF.
Causal confusion is a phenomenon where an agent learns a policy that reflects imperfect spurious correlations in the data. Such a policy may falsely appear to be optimal during training if most of the training data contain such spurious correlations. This phenomenon is particularly pronounced in domains such as robotics, with potentially large gaps between the open- and closed-loop performance of an agent. In such settings, causally confused models may appear to perform well according to open-loop metrics during training but fail catastrophically when deployed in the real world. In this paper, we study causal confusion in offline reinforcement learning. We investigate whether selectively sampling appropriate points from a dataset of demonstrations may enable offline reinforcement learning agents to disambiguate the underlying causal mechanisms of the environment, alleviate causal confusion in offline reinforcement learning, and produce a safer model for deployment. To answer this question, we consider a set of tailored offline reinforcement learning datasets that exhibit causal ambiguity and assess the ability of active sampling techniques to reduce causal confusion at evaluation. We provide empirical evidence that uniform and active sampling techniques are able to consistently reduce causal confusion as training progresses and that active sampling is able to do so significantly more efficiently than uniform sampling.
This paper proposes a novel logo image recognition approach incorporating a localization technique based on reinforcement learning. Logo recognition is an image classification task identifying a brand in an image. As the size and position of a logo vary widely from image to image, it is necessary to determine its position for accurate recognition. However, because there is no annotation for the position coordinates, it is impossible to train and infer the location of the logo in the image. Therefore, we propose a deep reinforcement learning localization method for logo recognition (RL-LOGO). It utilizes deep reinforcement learning to identify a logo region in images without annotations of the positions, thereby improving classification accuracy. We demonstrated a significant improvement in accuracy compared with existing methods in several published benchmarks. Specifically, we achieved an 18-point accuracy improvement over competitive methods on the complex dataset Logo-2K+. This demonstrates that the proposed method is a promising approach to logo recognition in real-world applications.
Existing methods provide varying algorithms for different types of Boolean satisfiability problems (SAT), lacking a general solution framework. Accordingly, this study proposes a unified framework DCSAT based on integer programming and reinforcement learning (RL) algorithm to solve different types of SAT problems such as MaxSAT, Weighted MaxSAT, PMS, WPMS. Specifically, we first construct a consolidated integer programming representation for four types of SAT problems by adjusting objective function coefficients. Secondly, we construct an appropriate reinforcement learning models based on the 0-1 integer programming for SAT problems. Based on the binary tree search structure, we apply the Monte Carlo tree search (MCTS) method on SAT problems. Finally, we prove that this method can find all optimal Boolean assignments based on Wiener-khinchin law of large Numbers. We experimentally verify that this paradigm can prune the unnecessary search space to find the optimal Boolean assignments for the problem. Furthermore, the proposed method can provide diverse labels for supervised learning methods for SAT problems.