In this paper, we present a framework to understand the convergence of commonly used Q-learning reinforcement learning algorithms in practice. Two salient features of such algorithms are: (i)~the Q-table is recursively updated using an agent state (such as the state of a recurrent neural network) which is not a belief state or an information state and (ii)~policy regularization is often used to encourage exploration and stabilize the learning algorithm. We investigate the simplest form of such Q-learning algorithms which we call regularized agent-state-based Q-learning (RASQL) and show that it converges under mild technical conditions to the fixed point of an appropriately defined regularized MDP, which depends on the stationary distribution induced by the behavioral policy. We also show that a similar analysis continues to work for a variant of RASQL that learns periodic policies. We present numerical examples to illustrate that the empirical convergence behavior matches with the proposed theoretical limit.

Sequential decision-making is central to sustainable agricultural management and precision agriculture, where resource inputs must be optimized under uncertainty and over time. However, such decisions must often be made with limited observations, whereas classical bandit and reinforcement learning approaches typically rely on either linear or black-box reward models that may misrepresent domain knowledge or require large amounts of data. We propose a family of nonlinear, model-based bandit algorithms that embed domain-specific response curves directly into the exploration-exploitation loop. By coupling (i) principled uncertainty quantification with (ii) closed-form or rapidly computable profit optima, these algorithms achieve sublinear regret and near-optimal sample complexity while preserving interpretability. Theoretical analysis establishes regret and sample complexity bounds, and extensive simulations emulating real-world fertilizer-rate decisions show consistent improvements over both linear and nonparametric baselines (such as linear UCB and $k$-NN UCB) in the low-sample regime, under both well-specified and shape-compatible misspecified models. Because our approach leverages mechanistic insight rather than large data volumes, it is especially suited to resource-constrained settings, supporting sustainable, inclusive, and transparent sequential decision-making across agriculture, environmental management, and allied applications. This methodology directly contributes to SDG 2 (Zero Hunger) and SDG 12 (Responsible Consumption and Production) by enabling data-driven, less wasteful agricultural practices.

We study the problem of learning the optimal control policy for fine-tuning a given diffusion process, using general value function approximation. We develop a new class of algorithms by solving a variational inequality problem based on the Hamilton-Jacobi-Bellman (HJB) equations. We prove sharp statistical rates for the learned value function and control policy, depending on the complexity and approximation errors of the function class. In contrast to generic reinforcement learning problems, our approach shows that fine-tuning can be achieved via supervised regression, with faster statistical rate guarantees.

Notifications are an important communication channel for delivering timely and relevant information. Optimizing their delivery involves addressing complex sequential decision-making challenges under constraints such as message utility and user fatigue. Offline reinforcement learning (RL) methods, such as Conservative Q-Learning (CQL), have been applied to this problem but face practical challenges at scale, including instability, sensitivity to distribution shifts, limited reproducibility, and difficulties with explainability in high-dimensional recommendation settings. We present a Decision Transformer (DT) based framework that reframes policy learning as return-conditioned supervised learning, improving robustness, scalability, and modeling flexibility. Our contributions include a real-world comparison with CQL, a multi-reward design suitable for non-episodic tasks, a quantile regression approach to return-to-go conditioning, and a production-ready system with circular buffer-based sequence processing for near-real-time inference. Extensive offline and online experiments in a deployed notification system show that our approach improves notification utility and overall session activity while minimizing user fatigue. Compared to a multi-objective CQL-based agent, the DT-based approach achieved a +0.72% increase in sessions for notification decision-making at LinkedIn by making notification recommendation more relevant.

Bearing faults in rotating machinery can lead to significant operational disruptions and maintenance costs. Modern methods for bearing fault diagnosis rely heavily on vibration analysis and machine learning techniques, which often require extensive labeled data and may not adapt well to dynamic environments. This study explores the feasibility of reinforcement learning (RL), specifically Deep Q-Networks (DQNs), for bearing fault classification tasks in machine condition monitoring to enhance the accuracy and adaptability of bearing fault diagnosis. The results demonstrate that while RL models developed in this study can match the performance of traditional supervised learning models under controlled conditions, they excel in adaptability when equipped with optimized reward structures. However, their computational demands highlight areas for further improvement. These findings demonstrate RL's potential to complement traditional methods, paving the way for adaptive diagnostic frameworks.

Humans naturally exhibit bilateral symmetry in their gross manipulation skills, effortlessly mirroring simple actions between left and right hands. Bimanual robots-which also feature bilateral symmetry-should similarly exploit this property to perform tasks with either hand. Unlike humans, who often favor a dominant hand for fine dexterous skills, robots should ideally execute ambidextrous manipulation with equal proficiency. To this end, we introduce SYMDEX (SYMmetric DEXterity), a reinforcement learning framework for ambidextrous bi-manipulation that leverages the robot's inherent bilateral symmetry as an inductive bias. SYMDEX decomposes complex bimanual manipulation tasks into per-hand subtasks and trains dedicated policies for each. By exploiting bilateral symmetry via equivariant neural networks, experience from one arm is inherently leveraged by the opposite arm. We then distill the subtask policies into a global ambidextrous policy that is independent of the hand-task assignment. We evaluate SYMDEX on six challenging simulated manipulation tasks and demonstrate successful real-world deployment on two of them. Our approach strongly outperforms baselines on complex task in which the left and right hands perform different roles. We further demonstrate SYMDEX's scalability by extending it to a four-arm manipulation setup, where our symmetry-aware policies enable effective multi-arm collaboration and coordination. Our results highlight how structural symmetry as inductive bias in policy learning enhances sample efficiency, robustness, and generalization across diverse dexterous manipulation tasks.

Resource allocation plays a critical role in minimizing cycle time and improving the efficiency of business processes. Recently, Deep Reinforcement Learning (DRL) has emerged as a powerful technique to optimize resource allocation policies in business processes. In the DRL framework, an agent learns a policy through interaction with the environment, guided solely by reward signals that indicate the quality of its decisions. However, existing algorithms are not suitable for dynamic environments such as business processes. Furthermore, existing DRL-based methods rely on engineered reward functions that approximate the desired objective, but a misalignment between reward and objective can lead to undesired decisions or suboptimal policies. To address these issues, we propose a rollout-based DRL algorithm and a reward function to optimize the objective directly. Our algorithm iteratively improves the policy by evaluating execution trajectories following different actions. Our reward function directly decomposes the objective function of minimizing the cycle time, such that trial-and-error reward engineering becomes unnecessary. We evaluated our method in six scenarios, for which the optimal policy can be computed, and on a set of increasingly complex, realistically sized process models. The results show that our algorithm can learn the optimal policy for the scenarios and outperform or match the best heuristics on the realistically sized business processes.

Recent advances in applying reinforcement learning (RL) to large language models (LLMs) have led to substantial progress. In particular, a series of remarkable yet often counterintuitive phenomena have been reported in LLMs, exhibiting patterns not typically observed in traditional RL settings. For example, notable claims include that a single training example can match the performance achieved with an entire dataset, that the reward signal does not need to be very accurate, and that training solely with negative samples can match or even surpass sophisticated reward-based methods. However, the precise conditions under which these observations hold - and, critically, when they fail - remain unclear. In this work, we identify a key factor that differentiates RL observations: whether the pretrained model already exhibits strong Model-Task Alignment, as measured by pass@k accuracy on the evaluated task. Through a systematic and comprehensive examination of a series of counterintuitive claims, supported by rigorous experimental validation across different model architectures and task domains, our findings show that while standard RL training remains consistently robust across settings, many of these counterintuitive results arise only when the model and task already exhibit strong model-task alignment. In contrast, these techniques fail to drive substantial learning in more challenging regimes, where standard RL methods remain effective.

This paper develops a semi-on-demand transit feeder service using shared autonomous vehicles (SAVs) and zonal dispatching control based on reinforcement learning (RL). This service combines the cost-effectiveness of fixed-route transit with the adaptability of demand-responsive transport to improve accessibility in lower-density areas. Departing from the terminus, SAVs first make scheduled fixed stops, then offer on-demand pick-ups and drop-offs in a pre-determined flexible-route area. Our deep RL model dynamically assigns vehicles to subdivided flexible-route zones in response to real-time demand fluctuations and operations, using a policy gradient algorithm - Proximal Policy Optimization. The methodology is demonstrated through agent-based simulations on a real-world bus route in Munich, Germany. Results show that after efficient training of the RL model, the semi-on-demand service with dynamic zonal control serves 16% more passengers at 13% higher generalized costs on average compared to traditional fixed-route service. The efficiency gain brought by RL control brings 2.4% more passengers at 1.4% higher costs. This study not only showcases the potential of integrating SAV feeders and machine learning techniques into public transit, but also sets the groundwork for further innovations in addressing first-mile-last-mile problems in multimodal transit systems.

Efficient robot control often requires balancing task performance with energy expenditure. A common approach in reinforcement learning (RL) is to penalize energy use directly as part of the reward function. This requires carefully tuning weight terms to avoid undesirable trade-offs where energy minimization harms task success. In this work, we propose a hyperparameter-free gradient optimization method to minimize energy expenditure without conflicting with task performance. Inspired by recent works in multitask learning, our method applies policy gradient projection between task and energy objectives to derive policy updates that minimize energy expenditure in ways that do not impact task performance. We evaluate this technique on standard locomotion benchmarks of DM-Control and HumanoidBench and demonstrate a reduction of 64% energy usage while maintaining comparable task performance. Further, we conduct experiments on a Unitree GO2 quadruped showcasing Sim2Real transfer of energy efficient policies. Our method is easy to implement in standard RL pipelines with minimal code changes, is applicable to any policy gradient method, and offers a principled alternative to reward shaping for energy efficient control policies.