Policy Mirror Descent (PMD) has emerged as a unifying framework in reinforcement learning (RL) by linking policy gradient methods with a first-order optimization method known as mirror descent. At its core, PMD incorporates two key regularization components: (i) a distance term that enforces a trust region for stable policy updates and (ii) an MDP regularizer that augments the reward function to promote structure and robustness. While PMD has been extensively studied in theory, empirical investigations remain scarce. This work provides a large-scale empirical analysis of the interplay between these two regularization techniques, running over 500k training seeds on small RL environments. Our results demonstrate that, although the two regularizers can partially substitute each other, their precise combination is critical for achieving robust performance.

Inverse Reinforcement Learning (IRL) presents a powerful paradigm for learning complex robotic tasks from human demonstrations. However, most approaches make the assumption that expert demonstrations are available, which is often not the case. Those that allow for suboptimality in the demonstrations are not designed for long-horizon goals or adversarial tasks. Many desirable robot capabilities fall into one or both of these categories, thus highlighting a critical shortcoming in the ability of IRL to produce field-ready robotic agents. We introduce Sample-efficient Preference-based inverse reinforcement learning for Long-horizon Adversarial tasks from Suboptimal Hierarchical demonstrations (SPLASH), which advances the state-of-the-art in learning from suboptimal demonstrations to long-horizon and adversarial settings. We empirically validate SPLASH on a maritime capture-the-flag task in simulation, and demonstrate real-world applicability with sim-to-real translation experiments on autonomous unmanned surface vehicles. We show that our proposed methods allow SPLASH to significantly outperform the state-of-the-art in reward learning from suboptimal demonstrations.

We present elsciRL, an open-source Python library to facilitate the application of language solutions on reinforcement learning problems. We demonstrate the potential of our software by extending the Language Adapter with Self-Completing Instruction framework defined in (Osborne, 2024) with the use of LLMs. Our approach can be re-applied to new applications with minimal setup requirements. We provide a novel GUI that allows a user to provide text input for an LLM to generate instructions which it can then self-complete. Empirical results indicate that these instructions \textit{can} improve a reinforcement learning agent's performance. Therefore, we present this work to accelerate the evaluation of language solutions on reward based environments to enable new opportunities for scientific discovery.

In Wireless Networked Control Systems (WNCSs), control and communication systems must be co-designed due to their strong interdependence. This paper presents a novel optimization theory-based safe deep reinforcement learning (DRL) framework for ultra-reliable WNCSs, ensuring constraint satisfaction while optimizing performance, for the first time in the literature. The approach minimizes power consumption under key constraints, including Peak Age of Information (PAoI) violation probability, transmit power, and schedulability in the finite blocklength regime. PAoI violation probability is uniquely derived by combining stochastic maximum allowable transfer interval (MATI) and maximum allowable packet delay (MAD) constraints in a multi-sensor network. The framework consists of two stages: optimization theory and safe DRL. The first stage derives optimality conditions to establish mathematical relationships among variables, simplifying and decomposing the problem. The second stage employs a safe DRL model where a teacher-student framework guides the DRL agent (student). The control mechanism (teacher) evaluates compliance with system constraints and suggests the nearest feasible action when needed. Extensive simulations show that the proposed framework outperforms rule-based and other optimization theory based DRL benchmarks, achieving faster convergence, higher rewards, and greater stability.

Image captioning is an important problem in developing various AI systems, and these tasks require large volumes of annotated images to train the models. Since all existing labelled datasets are already used for training the large Vision Language Models (VLMs), it becomes challenging to improve the performance of the same. Considering this, it is essential to consider the unsupervised image captioning performance, which remains relatively under-explored. To that end, we propose LoGIC (Lewis Communication Game for Image Captioning), a Multi-agent Reinforcement Learning game. The proposed method consists of two agents, a 'speaker' and a 'listener', with the objective of learning a strategy for communicating in natural language. We train agents in the cooperative common-reward setting using the GRPO algorithm and show that improvement in image captioning performance emerges as a consequence of the agents learning to play the game. We show that using pre-trained VLMs as the 'speaker' and Large Language Model (LLM) for language understanding in the 'listener', we achieved a $46$ BLEU score after fine-tuning using LoGIC without additional labels, a $2$ units advantage in absolute metrics compared to the $44$ BLEU score of the vanilla VLM. Additionally, we replace the VLM from the 'speaker' with lightweight components: (i) a ViT for image perception and (ii) a GPT2 language generation, and train them from scratch using LoGIC, obtaining a $31$ BLEU score in the unsupervised setting, a $10$ points advantage over existing unsupervised image-captioning methods.

Segment Anything Model 2 (SAM 2) has demonstrated strong performance in object segmentation tasks and has become the state-of-the-art for visual object tracking. The model stores information from previous frames in a memory bank, enabling temporal consistency across video sequences. Recent methods augment SAM 2 with hand-crafted update rules to better handle distractors, occlusions, and object motion. We propose a fundamentally different approach using reinforcement learning for optimizing memory updates in SAM 2 by framing memory control as a sequential decision-making problem. In an overfitting setup with a separate agent per video, our method achieves a relative improvement over SAM 2 that exceeds by more than three times the gains of existing heuristics. These results reveal the untapped potential of the memory bank and highlight reinforcement learning as a powerful alternative to hand-crafted update rules for memory control in visual object tracking.

Offline-to-online reinforcement learning (RL) aims to integrate the complementary strengths of offline and online RL by pre-training an agent offline and subsequently fine-tuning it through online interactions. However, recent studies reveal that offline pre-trained agents often underperform during online fine-tuning due to inaccurate value estimation caused by distribution shift, with random initialization proving more effective in certain cases. In this work, we propose a novel method, Online Pre-Training for Offline-to-Online RL (OPT), explicitly designed to address the issue of inaccurate value estimation in offline pre-trained agents. OPT introduces a new learning phase, Online Pre-Training, which allows the training of a new value function tailored specifically for effective online fine-tuning. Implementation of OPT on TD3 and SPOT demonstrates an average 30% improvement in performance across a wide range of D4RL environments, including MuJoCo, Antmaze, and Adroit.

Reliable satellite attitude control is essential for the success of space missions, particularly as satellites increasingly operate autonomously in dynamic and uncertain environments. Reaction wheels (RWs) play a pivotal role in attitude control, and maintaining control resilience during RW faults is critical to preserving mission objectives and system stability. However, traditional Proportional Derivative (PD) controllers and existing deep reinforcement learning (DRL) algorithms such as TD3, PPO, and A2C often fall short in providing the real time adaptability and fault tolerance required for autonomous satellite operations. This study introduces a DRL-based control strategy designed to improve satellite resilience and adaptability under fault conditions. Specifically, the proposed method integrates Twin Delayed Deep Deterministic Policy Gradient (TD3) with Hindsight Experience Replay (HER) and Dimension Wise Clipping (DWC) referred to as TD3-HD to enhance learning in sparse reward environments and maintain satellite stability during RW failures. The proposed approach is benchmarked against PD control and leading DRL algorithms. Experimental results show that TD3-HD achieves significantly lower attitude error, improved angular velocity regulation, and enhanced stability under fault conditions. These findings underscore the proposed method potential as a powerful, fault tolerant, onboard AI solution for autonomous satellite attitude control.

What drives an agent to explore the world while also maintaining control over the environment? From a child at play to scientists in the lab, intelligent agents must balance curiosity (the drive to seek knowledge) with competence (the drive to master and control the environment). Bridging cognitive theories of intrinsic motivation with reinforcement learning, we ask how evolving internal representations mediate the trade-off between curiosity (novelty or information gain) and competence (empowerment). We compare two model-based agents using handcrafted state abstractions (Tabular) or learning an internal world model (Dreamer). The Tabular agent shows curiosity and competence guide exploration in distinct patterns, while prioritizing both improves exploration. The Dreamer agent reveals a two-way interaction between exploration and representation learning, mirroring the developmental co-evolution of curiosity and competence.

In most real-world applications of artificial intelligence, the distributions of the data and the goals of the learners tend to change over time. The Probably Approximately Correct (PAC) learning framework, which underpins most machine learning algorithms, fails to account for dynamic data distributions and evolving objectives, often resulting in suboptimal performance. Prospective learning is a recently introduced mathematical framework that overcomes some of these limitations. We build on this framework to present preliminary results that improve the algorithm and numerical results, and extend prospective learning to sequential decision-making scenarios, specifically foraging. Code is available at: https://github.com/neurodata/prolearn2.