We introduce UnrealZoo, a rich collection of photo-realistic 3D virtual worlds built on Unreal Engine, designed to reflect the complexity and variability of the open worlds. Additionally, we offer a variety of playable entities for embodied AI agents. Based on UnrealCV, we provide a suite of easy-to-use Python APIs and tools for various potential applications, such as data collection, environment augmentation, distributed training, and benchmarking. We optimize the rendering and communication efficiency of UnrealCV to support advanced applications, such as multi-agent interaction. Our experiments benchmark agents in various complex scenes, focusing on visual navigation and tracking, which are fundamental capabilities for embodied visual intelligence. The results yield valuable insights into the advantages of diverse training environments for reinforcement learning (RL) agents and the challenges faced by current embodied vision agents, including those based on RL and large vision-language models (VLMs), in open worlds. These challenges involve latency in closed-loop control in dynamic scenes and reasoning about 3D spatial structures in unstructured terrain.

Exploration in cooperative multi-agent reinforcement learning (MARL) remains challenging for value-based agents due to the absence of an explicit policy. Existing approaches include individual exploration based on uncertainty towards the system and collective exploration through behavioral diversity among agents. However, the introduction of additional structures often leads to reduced training efficiency and infeasible integration of these methods. In this paper, we propose Adaptive exploration via Identity Recognition (AIR), which consists of two adversarial components: a classifier that recognizes agent identities from their trajectories, and an action selector that adaptively adjusts the mode and degree of exploration. We theoretically prove that AIR can facilitate both individual and collective exploration during training, and experiments also demonstrate the efficiency and effectiveness of AIR across various tasks.

We propose a decentralized, learning-based framework for dynamic coalition formation in Multi-Robot Task Allocation (MRTA). Our approach extends Multi-Agent Proximal Policy Optimization (MAPPO) by incorporating spatial action maps, robot motion control, task allocation revision, and intention sharing to enable effective coalition formation. Extensive simulations demonstrate that our model significantly outperforms existing methods, including a market-based baseline. Furthermore, we assess the scalability and generalizability of the proposed framework, highlighting its ability to handle large robot populations and adapt to diverse task allocation environments.

We present a survey paper on methods and applications of digital twins (DT) for urban traffic management. While the majority of studies on the DT focus on its "eyes," which is the emerging sensing and perception like object detection and tracking, what really distinguishes the DT from a traditional simulator lies in its ``brain," the prediction and decision making capabilities of extracting patterns and making informed decisions from what has been seen and perceived. In order to add values to urban transportation management, DTs need to be powered by artificial intelligence and complement with low-latency high-bandwidth sensing and networking technologies. We will first review the DT pipeline leveraging cyberphysical systems and propose our DT architecture deployed on a real-world testbed in New York City. This survey paper can be a pointer to help researchers and practitioners identify challenges and opportunities for the development of DTs; a bridge to initiate conversations across disciplines; and a road map to exploiting potentials of DTs for diverse urban transportation applications.

The exponential growth of data-intensive applications has placed unprecedented demands on modern storage systems, necessitating dynamic and efficient optimization strategies. Traditional heuristics employed for storage performance optimization often fail to adapt to the variability and complexity of contemporary workloads, leading to significant performance bottlenecks and resource inefficiencies. To address these challenges, this paper introduces RL-Storage, a novel reinforcement learning (RL)-based framework designed to dynamically optimize storage system configurations. RL-Storage leverages deep Q-learning algorithms to continuously learn from real-time I/O patterns and predict optimal storage parameters, such as cache size, queue depths, and readahead settings[1]. The proposed framework operates within the storage kernel, ensuring minimal latency and low computational overhead. Through an adaptive feedback mechanism, RL-Storage dynamically adjusts critical parameters, achieving efficient resource utilization across a wide range of workloads. Experimental evaluations conducted on a range of benchmarks, including RocksDB and PostgreSQL, demonstrate significant improvements, with throughput gains of up to 2.6x and latency reductions of 43% compared to baseline heuristics. Additionally, RL-Storage achieves these performance enhancements with a negligible CPU overhead of 0.11% and a memory footprint of only 5 KB, making it suitable for seamless deployment in production environments. This work underscores the transformative potential of reinforcement learning techniques in addressing the dynamic nature of modern storage systems. By autonomously adapting to workload variations in real time, RL-Storage provides a robust and scalable solution for optimizing storage performance, paving the way for next-generation intelligent storage infrastructures.

This work in the field of developmental cognitive robotics aims to devise a new domain bridging between reinforcement learning and imitation learning, with a model of the intrinsic motivation for learning agents to learn with guidance from tutors multiple tasks, including sequential tasks. The main contribution has been to propose a common formulation of intrinsic motivation based on empirical progress for a learning agent to choose automatically its learning curriculum by actively choosing its learning strategy for simple or sequential tasks: which task to learn, between autonomous exploration or imitation learning, between low-level actions or task decomposition, between several tutors. The originality is to design a learner that benefits not only passively from data provided by tutors, but to actively choose when to request tutoring and what and whom to ask. The learner is thus more robust to the quality of the tutoring and learns faster with fewer demonstrations. We developed the framework of socially guided intrinsic motivation with machine learning algorithms to learn multiple tasks by taking advantage of the generalisability properties of human demonstrations in a passive manner or in an active manner through requests of demonstrations from the best tutor for simple and composing subtasks. The latter relies on a representation of subtask composition proposed for a construction process, which should be refined by representations used for observational processes of analysing human movements and activities of daily living. With the outlook of a language-like communication with the tutor, we investigated the emergence of a symbolic representation of the continuous sensorimotor space and of tasks using intrinsic motivation. We proposed within the reinforcement learning framework, a reward function for interacting with tutors for automatic curriculum learning in multi-task learning.

Model-based reinforcement learning aims to increase sample efficiency, but the accuracy of dynamics models and the resulting compounding errors are often seen as key limitations. This paper empirically investigates potential sample efficiency gains from improved dynamics models in model-based value expansion methods. Our study reveals two key findings when using oracle dynamics models to eliminate compounding errors. First, longer rollout horizons enhance sample efficiency, but the improvements quickly diminish with each additional expansion step. Second, increased model accuracy only marginally improves sample efficiency compared to learned models with identical horizons. These diminishing returns in sample efficiency are particularly noteworthy when compared to model-free value expansion methods. These model-free algorithms achieve comparable performance without the computational overhead. Our results suggest that the limitation of model-based value expansion methods cannot be attributed to model accuracy. Although higher accuracy is beneficial, even perfect models do not provide unrivaled sample efficiency. Therefore, the bottleneck exists elsewhere. These results challenge the common assumption that model accuracy is the primary constraint in model-based reinforcement learning.

Offline reinforcement learning (RL) enables policy learning from pre-collected offline datasets, relaxing the need to interact directly with the environment. However, limited by the quality of offline datasets, it generally fails to learn well-qualified policies in suboptimal datasets. To address datasets with insufficient optimal demonstrations, we introduce Goal-cOnditioned Data Augmentation (GODA), a novel goal-conditioned diffusion-based method for augmenting samples with higher quality. Leveraging recent advancements in generative modeling, GODA incorporates a novel return-oriented goal condition with various selection mechanisms. Specifically, we introduce a controllable scaling technique to provide enhanced return-based guidance during data sampling. GODA learns a comprehensive distribution representation of the original offline datasets while generating new data with selectively higher-return goals, thereby maximizing the utility of limited optimal demonstrations. Furthermore, we propose a novel adaptive gated conditioning method for processing noised inputs and conditions, enhancing the capture of goal-oriented guidance. We conduct experiments on the D4RL benchmark and real-world challenges, specifically traffic signal control (TSC) tasks, to demonstrate GODA's effectiveness in enhancing data quality and superior performance compared to state-of-the-art data augmentation methods across various offline RL algorithms.

The high costs and risks involved in extensive environment interactions hinder the practical application of current online safe reinforcement learning (RL) methods. While offline safe RL addresses this by learning policies from static datasets, the performance therein is usually limited due to reliance on data quality and challenges with out-of-distribution (OOD) actions. Inspired by recent successes in offline-to-online (O2O) RL, it is crucial to explore whether offline safe RL can be leveraged to facilitate faster and safer online policy learning, a direction that has yet to be fully investigated. To fill this gap, we first demonstrate that naively applying existing O2O algorithms from standard RL would not work well in the safe RL setting due to two unique challenges: \emph{erroneous Q-estimations}, resulted from offline-online objective mismatch and offline cost sparsity, and \emph{Lagrangian mismatch}, resulted from difficulties in aligning Lagrange multipliers between offline and online policies. To address these challenges, we introduce \textbf{Marvel}, a novel framework for O2O safe RL, comprising two key components that work in concert: \emph{Value Pre-Alignment} to align the Q-functions with the underlying truth before online learning, and \emph{Adaptive PID Control} to effectively adjust the Lagrange multipliers during online finetuning. Extensive experiments demonstrate that Marvel significantly outperforms existing baselines in both reward maximization and safety constraint satisfaction. By introducing the first policy-finetuning based framework for O2O safe RL, which is compatible with many offline and online safe RL methods, our work has the great potential to advance the field towards more efficient and practical safe RL solutions.

The Cost-aware Dynamic Multi-Workflow Scheduling (CDMWS) in the cloud is a kind of cloud workflow management problem, which aims to assign virtual machine (VM) instances to execute tasks in workflows so as to minimize the total costs, including both the penalties for violating Service Level Agreement (SLA) and the VM rental fees. Powered by deep neural networks, Reinforcement Learning (RL) methods can construct effective scheduling policies for solving CDMWS problems. Traditional policy networks in RL often use basic feedforward architectures to separately determine the suitability of assigning any VM instances, without considering all VMs simultaneously to learn their global information. This paper proposes a novel self-attention policy network for cloud workflow scheduling (SPN-CWS) that captures global information from all VMs. We also develop an Evolution Strategy-based RL (ERL) system to train SPN-CWS reliably and effectively. The trained SPN-CWS can effectively process all candidate VM instances simultaneously to identify the most suitable VM instance to execute every workflow task. Comprehensive experiments show that our method can noticeably outperform several state-of-the-art algorithms on multiple benchmark CDMWS problems.