In recent years, \emph{Reinforcement Learning} (RL) has made remarkable progress, achieving superhuman performance in a wide range of simulated environments. As research moves toward deploying RL in real-world applications, the field faces a new set of challenges inherent to real-world settings, such as large state-action spaces, non-stationarity, and partial observability. Despite their importance, these challenges are often underexplored in current benchmarks, which tend to focus on idealized, fully observable, and stationary environments, often neglecting to incorporate real-world complexities explicitly. In this paper, we introduce \texttt{Gym4ReaL}, a comprehensive suite of realistic environments designed to support the development and evaluation of RL algorithms that can operate in real-world scenarios. The suite includes a diverse set of tasks that expose algorithms to a variety of practical challenges. Our experimental results show that, in these settings, standard RL algorithms confirm their competitiveness against rule-based benchmarks, motivating the development of new methods to fully exploit the potential of RL to tackle the complexities of real-world tasks.

The Cross-Entropy Method (CEM) is a widely adopted trajectory optimizer in model-based reinforcement learning (MBRL), but its unimodal sampling strategy often leads to premature convergence in multimodal landscapes. In this work, we propose Bregman Centroid Guided CEM ($\mathcal{BC}$-EvoCEM), a lightweight enhancement to ensemble CEM that leverages $\textit{Bregman centroids}$ for principled information aggregation and diversity control. $\textbf{$\mathcal{BC}$-EvoCEM}$ computes a performance-weighted Bregman centroid across CEM workers and updates the least contributing ones by sampling within a trust region around the centroid. Leveraging the duality between Bregman divergences and exponential family distributions, we show that $\textbf{$\mathcal{BC}$-EvoCEM}$ integrates seamlessly into standard CEM pipelines with negligible overhead. Empirical results on synthetic benchmarks, a cluttered navigation task, and full MBRL pipelines demonstrate that $\textbf{$\mathcal{BC}$-EvoCEM}$ enhances both convergence and solution quality, providing a simple yet effective upgrade for CEM.

Efficient elevator traffic management in large buildings is critical for minimizing passenger travel times and energy consumption. Because heuristic- or pattern-detection-based controllers struggle with the stochastic and combinatorial nature of dispatching, we model the six-elevator, fifteen-floor system at Vrije Universiteit Amsterdam as a Markov Decision Process and train an end-to-end Reinforcement Learning (RL) Elevator Group Control System (EGCS). Key innovations include a novel action space encoding to handle the combinatorial complexity of elevator dispatching, the introduction of infra-steps to model continuous passenger arrivals, and a tailored reward signal to improve learning efficiency. In addition, we explore various ways to adapt the discounting factor to the infra-step formulation. We investigate RL architectures based on Dueling Double Deep Q-learning, showing that the proposed RL-based EGCS adapts to fluctuating traffic patterns, learns from a highly stochastic environment, and thereby outperforms a traditional rule-based algorithm.

Overestimation is pervasive in reinforcement learning (RL), including in Q-learning, which forms the algorithmic basis for many value-based deep RL algorithms. Double Q-learning is an algorithm introduced to address Q-learning's overestimation by training two Q-functions and using both to de-correlate action-selection and action-evaluation in bootstrap targets. Shortly after Q-learning was adapted to deep RL in the form of deep Q-networks (DQN), Double Q-learning was adapted to deep RL in the form of Double DQN. However, Double DQN only loosely adapts Double Q-learning, forgoing the training of two different Q-functions that bootstrap off one another. In this paper, we study algorithms that adapt this core idea of Double Q-learning for value-based deep RL. We term such algorithms Deep Double Q-learning (DDQL). Our aim is to understand whether DDQL exhibits less overestimation than Double DQN and whether performant instantiations of DDQL exist. We answer both questions affirmatively, demonstrating that DDQL reduces overestimation and outperforms Double DQN in aggregate across 57 Atari 2600 games, without requiring additional hyperparameters. We also study several aspects of DDQL, including its network architecture, replay ratio, and minibatch sampling strategy.

Reinforcement learning (RL) enables agents to learn optimal policies through environmental interaction. However, RL suffers from reduced learning efficiency due to the curse of dimensionality in high-dimensional spaces. Quantum reinforcement learning (QRL) addresses this issue by leveraging superposition and entanglement in quantum computing, allowing efficient handling of high-dimensional problems with fewer resources. QRL combines quantum neural networks (QNNs) with RL, where the parameterized quantum circuit (PQC) acts as the core computational module. The PQC performs linear and nonlinear transformations through gate operations, similar to hidden layers in classical neural networks. Previous QRL studies, however, have used fixed PQC structures based on empirical intuition without verifying their optimality. This paper proposes a QRL-NAS algorithm that integrates quantum neural architecture search (QNAS) to optimize PQC structures within QRL. Experiments demonstrate that QRL-NAS achieves higher rewards than QRL with fixed circuits, validating its effectiveness and practical utility.

Modernizing legacy software systems is a critical but challenging task, often hampered by a lack of documentation and understanding of the original system's intricate decision logic. Traditional approaches like behavioral cloning merely replicate input-output behavior without capturing the underlying intent. This paper proposes a novel pipeline to automatically extract interpretable decision logic from legacy systems treated as black boxes. The approach uses a Reinforcement Learning (RL) agent to explore the input space and identify critical decision boundaries by rewarding actions that cause meaningful changes in the system's output. These counterfactual state transitions, where the output changes, are collected and clustered using K-Means. Decision trees are then trained on these clusters to extract human-readable rules that approximate the system's decision logic near the identified boundaries. I demonstrated the pipeline's effectiveness on three dummy legacy systems with varying complexity, including threshold-based, combined-conditional, and non-linear range logic. Results show that the RL agent successfully focuses exploration on relevant boundary regions, and the extracted rules accurately reflect the core logic of the underlying dummy systems, providing a promising foundation for generating specifications and test cases during legacy migration.

Deep reinforcement learning (DRL) has become a powerful tool for complex decision-making in machine learning and AI. However, traditional methods often assume perfect action execution, overlooking the uncertainties and deviations between an agent's selected actions and the actual system response. In real-world applications, such as robotics, mechatronics, and communication networks, execution mismatches arising from system dynamics, hardware constraints, and latency can significantly degrade performance. This work advances AI by developing a novel control-optimized DRL framework that explicitly models and compensates for action execution mismatches, a challenge largely overlooked in existing methods. Our approach establishes a structured two-stage process: determining the desired action and selecting the appropriate control signal to ensure proper execution. It trains the agent while accounting for action mismatches and controller corrections. By incorporating these factors into the training process, the AI agent optimizes the desired action with respect to both the actual control signal and the intended outcome, explicitly considering execution errors. This approach enhances robustness, ensuring that decision-making remains effective under real-world uncertainties. Our approach offers a substantial advancement for engineering practice by bridging the gap between idealized learning and real-world implementation. It equips intelligent agents operating in engineering environments with the ability to anticipate and adjust for actuation errors and system disturbances during training. We evaluate the framework in five widely used open-source mechanical simulation environments we restructured and developed to reflect real-world operating conditions, showcasing its robustness against uncertainties and offering a highly practical and efficient solution for control-oriented applications.

Preference-based Reinforcement Learning (PbRL) provides a way to learn high-performance policies in environments where the reward signal is hard to specify, avoiding heuristic and time-consuming reward design. However, PbRL can suffer from slow convergence speed since it requires training in a reward model. Prior work has proposed learning a reward model from demonstrations and fine-tuning it using preferences. However, when the model is a neural network, using different loss functions for pre-training and fine-tuning can pose challenges to reliable optimization. In this paper, we propose a method to effectively leverage prior knowledge with a Residual Reward Model (RRM). An RRM assumes that the true reward of the environment can be split into a sum of two parts: a prior reward and a learned reward. The prior reward is a term available before training, for example, a user's ``best guess'' reward function, or a reward function learned from inverse reinforcement learning (IRL), and the learned reward is trained with preferences. We introduce state-based and image-based versions of RRM and evaluate them on several tasks in the Meta-World environment suite. Experimental results show that our method substantially improves the performance of a common PbRL method. Our method achieves performance improvements for a variety of different types of prior rewards, including proxy rewards, a reward obtained from IRL, and even a negated version of the proxy reward. We also conduct experiments with a Franka Panda to show that our method leads to superior performance on a real robot. It significantly accelerates policy learning for different tasks, achieving success in fewer steps than the baseline. The videos are presented at https://sunlighted.github.io/RRM-web/.

Sampling algorithms drive probabilistic machine learning, and recent years have seen an explosion in the diversity of tools for this task. However, the increasing sophistication of sampling algorithms is correlated with an increase in the tuning burden. There is now a greater need than ever to treat the tuning of samplers as a learning task in its own right. In a conceptual breakthrough, Wang et al (2025) formulated Metropolis-Hastings as a Markov decision process, opening up the possibility for adaptive tuning using Reinforcement Learning (RL). Their emphasis was on theoretical foundations; realising the practical benefit of Reinforcement Learning Metropolis-Hastings (RLMH) was left for subsequent work. The purpose of this paper is twofold: First, we observe the surprising result that natural choices of reward, such as the acceptance rate, or the expected squared jump distance, provide insufficient signal for training RLMH. Instead, we propose a novel reward based on the contrastive divergence, whose superior performance in the context of RLMH is demonstrated. Second, we explore the potential of RLMH and present adaptive gradient-based samplers that balance flexibility of the Markov transition kernel with learnability of the associated RL task. A comprehensive simulation study using the posteriordb benchmark supports the practical effectiveness of RLMH.

We propose Fractional Policy Gradients (FPG), a reinforcement learning framework incorporating fractional calculus for long-term temporal modeling in policy optimization. Standard policy gradient approaches face limitations from Markovian assumptions, exhibiting high variance and inefficient sampling. By reformulating gradients using Caputo fractional derivatives, FPG establishes power-law temporal correlations between state transitions. We develop an efficient recursive computation technique for fractional temporal-difference errors with constant time and memory requirements. Theoretical analysis shows FPG achieves asymptotic variance reduction of order O(t^(-alpha)) versus standard policy gradients while preserving convergence. Empirical validation demonstrates 35-68% sample efficiency gains and 24-52% variance reduction versus state-of-the-art baselines. This framework provides a mathematically grounded approach for leveraging long-range dependencies without computational overhead.