Abstract--Deep Reinforcement Learning (DRL) has emerged as an efficient approach to resource allocation due to its strong capability in handling complex decision-making tasks. However, only limited research has explored the training of DRL models with real-world data in practical, distributed Internet of Things (IoT) systems. T o bridge this gap, this paper proposes a novel framework for training DRL models in real-world distributed IoT environments. In the proposed framework, IoT devices select communication channels using a DRL-based method, while the DRL model is trained with feedback information--specifically, Acknowledgment (ACK) information--obtained from actual data transmissions over the selected channels. Implementation and performance evaluation, in terms of Frame Success Rate (FSR), are carried out, demonstrating both the feasibility and the effectiveness of the proposed framework. In recent years, the number of Internet of Things (IoT) devices has grown rapidly, driven by advancements in communication technologies such as LoRa, Sigfox, and NB-IoT, the declining cost of sensors and embedded systems, and the application of artificial intelligence in data processing.

The growing threat of low-cost kamikaze drone swarms poses a critical challenge to modern defense systems demanding rapid and strategic decision-making to prioritize interceptions across multiple effectors and high-value target zones. In this work, we present a case study demonstrating the practical advantages of reinforcement learning in addressing this challenge. We introduce a high-fidelity simulation environment that captures realistic operational constraints, within which a decision-level reinforcement learning agent learns to coordinate multiple effectors for optimal interception prioritization. Operating in a discrete action space, the agent selects which drone to engage per effector based on observed state features such as positions, classes, and effector status. We evaluate the learned policy against a handcrafted rule-based baseline across hundreds of simulated attack scenarios. The reinforcement learning based policy consistently achieves lower average damage and higher defensive efficiency in protecting critical zones. This case study highlights the potential of reinforcement learning as a strategic layer within defense architectures, enhancing resilience without displacing existing control systems. All code and simulation assets are publicly released for full reproducibility, and a video demonstration illustrates the policy's qualitative behavior.

Bisimulation metric has long been regarded as an effective control-related representation learning technique in various reinforcement learning tasks. However, in this paper, we identify two main issues with the conventional bisimulation metric: 1) an inability to represent certain distinctive scenarios, and 2) a reliance on predefined weights for differences in rewards and subsequent states during recursive updates. We find that the first issue arises from an imprecise definition of the reward gap, whereas the second issue stems from overlooking the varying importance of reward difference and next-state distinctions across different training stages and task settings. To address these issues, by introducing a measure for state-action pairs, we propose a revised bisimulation metric that features a more precise definition of reward gap and novel update operators with adaptive coefficient. We also offer theoretical guarantees of convergence for our proposed metric and its improved representation distinctiveness. In addition to our rigorous theoretical analysis, we conduct extensive experiments on two representative benchmarks, DeepMind Control and Meta-World, demonstrating the effectiveness of our approach.

Learning visuomotor policies via imitation has proven effective across a wide range of robotic domains. However, the performance of these policies is heavily dependent on the number of training demonstrations, which requires expensive data collection in the real world. In this work, we aim to reduce data collection efforts when learning visuomotor robot policies by leveraging existing or cost-effective data from a wide range of embodiments, such as public robot datasets and the datasets of humans playing with objects (human data from play). Our approach leverages two key insights. First, we use optic flow as an embodiment-agnostic action representation to train a World Model (WM) across multi-embodiment datasets, and finetune it on a small amount of robot data from the target embodiment. Second, we develop a method, Latent Policy Steering (LPS), to improve the output of a behavior-cloned policy by searching in the latent space of the WM for better action sequences. In real world experiments, we observe significant improvements in the performance of policies trained with a small amount of data (over 50% relative improvement with 30 demonstrations and over 20% relative improvement with 50 demonstrations) by combining the policy with a WM pretrained on two thousand episodes sampled from the existing Open X-embodiment dataset across different robots or a cost-effective human dataset from play.

Proficient game agents with diverse play styles enrich the gaming experience and enhance the replay value of games. However, recent advancements in game AI based on reinforcement learning have predominantly focused on improving proficiency, whereas methods based on evolution algorithms generate agents with diverse play styles but exhibit subpar performance compared to RL methods. To address this gap, this paper proposes Mixed Proximal Policy Optimization (MPPO), a method designed to improve the proficiency of existing suboptimal agents while retaining their distinct styles. MPPO unifies loss objectives for both online and offline samples and introduces an implicit constraint to approximate demonstrator policies by adjusting the empirical distribution of samples. Empirical results across environments of varying scales demonstrate that MPPO achieves proficiency levels comparable to, or even superior to, pure online algorithms while preserving demonstrators' play styles. This work presents an effective approach for generating highly proficient and diverse game agents, ultimately contributing to more engaging gameplay experiences.

Vehicle Routing Problems (VRP) are an extension of the Traveling Salesperson Problem and are a fundamental NP - hard challenge in combinatorial optimization. Solving VRP in real - time at large scale has become critical in numerous applications, from growing markets like last - mile delivery to emerging use - cases like interactive logistics planning. In many applications, one has to repeatedly solv e VRP instances dr a wn from the same distribution, yet current state - of - the - art solvers treat each instance on its own without leveraging previous examples . We introduce a n optimization framework where a reinforcement learning agent is trained on prior instances and quickly generate s initial solutions, which are then further optimized by a genetic algorithm. This framework, Evolutionary Algorithm with Reinforcement Learning Initialization ( EARLI), consistently outperforms current state - of - the - art solvers across various time budgets . For example, EARLI handles vehicle routing with 500 locations within one second, 10x faster than current solvers for the same solution quality, enabling real - time and interactive routing at scale . EARLI can generalize to new data, as we demonstrate on real e - commerce delivery data of a previously unseen city . By combin ing reinforcement learning and genetic algorithms, o ur hybrid framework takes a step forward to closer interdisciplinary collaboration between AI and optimization communities towards real - time optimization in diverse domains .

Contextual bandit algorithms have transformed modern experimentation by enabling real-time adaptation for personalized treatment and efficient use of data. Yet these advantages create challenges for statistical inference due to adaptivity. A fundamental property that supports valid inference is policy convergence, meaning that action-selection probabilities converge in probability given the context. Convergence ensures replicability of adaptive experiments and stability of online algorithms. In this paper, we highlight a previously overlooked issue: widely used algorithms such as LinUCB may fail to converge when the reward model is misspecified, and such non-convergence creates fundamental obstacles for statistical inference. This issue is practically important, as misspecified models -- such as linear approximations of complex dynamic system -- are often employed in real-world adaptive experiments to balance bias and variance. Motivated by this insight, we propose and analyze a broad class of algorithms that are guaranteed to converge even under model misspecification. Building on this guarantee, we develop a general inference framework based on an inverse-probability-weighted Z-estimator (IPW-Z) and establish its asymptotic normality with a consistent variance estimator. Simulation studies confirm that the proposed method provides robust and data-efficient confidence intervals, and can outperform existing approaches that exist only in the special case of offline policy evaluation. Taken together, our results underscore the importance of designing adaptive algorithms with built-in convergence guarantees to enable stable experimentation and valid statistical inference in practice.

As AI systems become increasingly autonomous, reliably aligning their decision-making with human preferences is essential. Inverse reinforcement learning (IRL) offers a promising approach to infer preferences from demonstrations. These preferences can then be used to produce an apprentice policy that performs well on the demonstrated task. However, in domains like autonomous driving or robotics, where errors can have serious consequences, we need not just good average performance but reliable policies with formal guarantees -- yet obtaining sufficient human demonstrations for reliability guarantees can be costly. Active IRL addresses this challenge by strategically selecting the most informative scenarios for human demonstration. We introduce PAC-EIG, an information-theoretic acquisition function that directly targets probably-approximately-correct (PAC) guarantees for the learned policy -- providing the first such theoretical guarantee for active IRL with noisy expert demonstrations. Our method maximises information gain about the regret of the apprentice policy, efficiently identifying states requiring further demonstration. We also present Reward-EIG as an alternative when learning the reward itself is the primary objective. Focusing on finite state-action spaces, we prove convergence bounds, illustrate failure modes of prior heuristic methods, and demonstrate our method's advantages experimentally.

In reinforcement learning with sparse rewards, demonstrations can accelerate learning, but determining when to imitate them remains challenging. We propose Smooth Policy Regularisation from Demonstrations (SPReD), a framework that addresses the fundamental question: when should an agent imitate a demonstration versus follow its own policy? SPReD uses ensemble methods to explicitly model Q-value distributions for both demonstration and policy actions, quantifying uncertainty for comparisons. We develop two complementary uncertainty-aware methods: a probabilistic approach estimating the likelihood of demonstration superiority, and an advantage-based approach scaling imitation by statistical significance. Unlike prevailing methods (e.g. Q-filter) that make binary imitation decisions, SPReD applies continuous, uncertainty-proportional regularisation weights, reducing gradient variance during training. Despite its computational simplicity, SPReD achieves remarkable gains in experiments across eight robotics tasks, outperforming existing approaches by up to a factor of 14 in complex tasks while maintaining robustness to demonstration quality and quantity. Our code is available at https://github.com/YujieZhu7/SPReD.

Online reinforcement learning (RL) has been central to post-training language models, but its extension to diffusion models remains challenging due to intractable likelihoods. Recent works discretize the reverse sampling process to enable GRPO-style training, yet they inherit fundamental drawbacks, including solver restrictions, forward-reverse inconsistency, and complicated integration with classifier-free guidance (CFG). We introduce Diffusion Negative-aware FineTuning (DiffusionNFT), a new online RL paradigm that optimizes diffusion models directly on the forward process via flow matching. DiffusionNFT contrasts positive and negative generations to define an implicit policy improvement direction, naturally incorporating reinforcement signals into the supervised learning objective. This formulation enables training with arbitrary black-box solvers, eliminates the need for likelihood estimation, and requires only clean images rather than sampling trajectories for policy optimization. DiffusionNFT is up to $25\times$ more efficient than FlowGRPO in head-to-head comparisons, while being CFG-free. For instance, DiffusionNFT improves the GenEval score from 0.24 to 0.98 within 1k steps, while FlowGRPO achieves 0.95 with over 5k steps and additional CFG employment. By leveraging multiple reward models, DiffusionNFT significantly boosts the performance of SD3.5-Medium in every benchmark tested.