We study online learning in episodic finite-horizon Markov decision processes (MDPs) with convex objective functions, known as the concave utility reinforcement learning (CURL) problem. This setting generalizes RL from linear to convex losses on the state-action distribution induced by the agent's policy. The non-linearity of CURL invalidates classical Bellman equations and requires new algorithmic approaches. We introduce the first algorithm achieving near-optimal regret bounds for online CURL without any prior knowledge on the transition function. To achieve this, we use an online mirror descent algorithm with varying constraint sets and a carefully designed exploration bonus. We then address for the first time a bandit version of CURL, where the only feedback is the value of the objective function on the state-action distribution induced by the agent's policy. We achieve a sub-linear regret bound for this more challenging problem by adapting techniques from bandit convex optimization to the MDP setting.

Many production lines require active control mechanisms, such as adaptive routing, worker reallocation, and rescheduling, to maintain optimal performance. However, designing these control systems is challenging for various reasons, and while reinforcement learning (RL) has shown promise in addressing these challenges, a standardized and general framework is still lacking. In this work, we introduce LineFlow, an extensible, open-source Python framework for simulating production lines of arbitrary complexity and training RL agents to control them. To demonstrate the capabilities and to validate the underlying theoretical assumptions of LineFlow, we formulate core subproblems of active line control in ways that facilitate mathematical analysis. For each problem, we provide optimal solutions for comparison. We benchmark state-of-the-art RL algorithms and show that the learned policies approach optimal performance in well-understood scenarios. However, for more complex, industrial-scale production lines, RL still faces significant challenges, highlighting the need for further research in areas such as reward shaping, curriculum learning, and hierarchical control.

We present an interpretability framework for unsupervised reinforcement learning (URL) agents, aimed at understanding how intrinsic motivation shapes attention, behavior, and representation learning. We analyze five agents DQN, RND, ICM, PPO, and a Transformer-RND variant trained on procedurally generated environments, using Grad-CAM, Layer-wise Relevance Propagation (LRP), exploration metrics, and latent space clustering. To capture how agents perceive and adapt over time, we introduce two metrics: attention diversity, which measures the spatial breadth of focus, and attention change rate, which quantifies temporal shifts in attention. Our findings show that curiosity-driven agents display broader, more dynamic attention and exploratory behavior than their extrinsically motivated counterparts. Among them, TransformerRND combines wide attention, high exploration coverage, and compact, structured latent representations. Our results highlight the influence of architectural inductive biases and training signals on internal agent dynamics. Beyond reward-centric evaluation, the proposed framework offers diagnostic tools to probe perception and abstraction in RL agents, enabling more interpretable and generalizable behavior.

Mobile crowdsensing is evolving beyond traditional human-centric models by integrating heterogeneous entities like unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs). Optimizing task allocation among these diverse agents is critical, particularly in challenging emergency rescue scenarios characterized by complex environments, limited communication, and partial observability. This paper tackles the Heterogeneous-Entity Collaborative-Sensing Task Allocation (HECTA) problem specifically for emergency rescue, considering humans, UAVs, and UGVs. We introduce a novel ``Hard-Cooperative'' policy where UGVs prioritize recharging low-battery UAVs, alongside performing their sensing tasks. The primary objective is maximizing the task completion rate (TCR) under strict time constraints. We rigorously formulate this NP-hard problem as a decentralized partially observable Markov decision process (Dec-POMDP) to effectively handle sequential decision-making under uncertainty. To solve this, we propose HECTA4ER, a novel multi-agent reinforcement learning algorithm built upon a Centralized Training with Decentralized Execution architecture. HECTA4ER incorporates tailored designs, including specialized modules for complex feature extraction, utilization of action-observation history via hidden states, and a mixing network integrating global and local information, specifically addressing the challenges of partial observability. Furthermore, theoretical analysis confirms the algorithm's convergence properties. Extensive simulations demonstrate that HECTA4ER significantly outperforms baseline algorithms, achieving an average 18.42% increase in TCR. Crucially, a real-world case study validates the algorithm's effectiveness and robustness in dynamic sensing scenarios, highlighting its strong potential for practical application in emergency response.

Generalist imitation learning policies trained on large datasets show great promise for solving diverse manipulation tasks. However, to ensure generalization to different conditions, policies need to be trained with data collected across a large set of environmental factor variations (e.g., camera pose, table height, distractors) $-$ a prohibitively expensive undertaking, if done exhaustively. We introduce a principled method for deciding what data to collect and how much to collect for each factor by constructing factored scaling curves (FSC), which quantify how policy performance varies as data scales along individual or paired factors. These curves enable targeted data acquisition for the most influential factor combinations within a given budget. We evaluate the proposed method through extensive simulated and real-world experiments, across both training-from-scratch and fine-tuning settings, and show that it boosts success rates in real-world tasks in new environments by up to 26% over existing data-collection strategies. We further demonstrate how factored scaling curves can effectively guide data collection using an offline metric, without requiring real-world evaluation at scale.

Industrial automation increasingly demands energy-efficient control strategies to balance performance with environmental and cost constraints. In this work, we present a multi-objective reinforcement learning (MORL) framework for energy-efficient control of the Quanser Aero 2 testbed in its one-degree-of-freedom configuration. We design a composite reward function that simultaneously penalizes tracking error and electrical power consumption. Preliminary experiments explore the influence of varying the Energy penalty weight, alpha, on the trade-off between pitch tracking and energy savings. Our results reveal a marked performance shift for alpha values between 0.0 and 0.25, with non-Pareto optimal solutions emerging at lower alpha values, on both the simulation and the real system. We hypothesize that these effects may be attributed to artifacts introduced by the adaptive behavior of the Adam optimizer, which could bias the learning process and favor bang-bang control strategies. Future work will focus on automating alpha selection through Gaussian Process-based Pareto front modeling and transitioning the approach from simulation to real-world deployment.

-- This report presents our reinforcement learning-based approach for the swing-up and stabilisation tasks of the acrobot and pendubot, tailored specifcially to the updated guidelines of the 3rd AI Olympics at ICRA 2025. Building upon our previously developed A verage-Reward Entropy Advantage Policy Optimization (AR-EAPO) algorithm, we refined our solution to effectively address the new competition scenarios and evaluation metrics. Extensive simulations validate that our controller robustly manages these revised tasks, demonstrating adaptability and effectiveness within the updated framework. Building upon prior competitions at IJCAI 2023 [3] and IROS 2024 [4], the current edition places particular emphasis on global policy robustness, requiring solutions for reliable swing-up stabilisation tasks from arbitrary initial configurations under significantly increased external disturbances. The competition maintains its use of two different configurations: the acrobot, characterised by an inactive shoulder joint, and the pendubot, with an inactive elbow joint.

Vision-Language-Action (VLA) models have shown great potential in general robotic decision-making tasks via imitation learning. However, the variable quality of training data often constrains the performance of these models. On the other hand, offline Reinforcement Learning (RL) excels at learning robust policy models from mixed-quality data. In this paper, we introduce Reinforced robot GPT (ReinboT), a novel end-to-end VLA model that integrates the RL principle of maximizing cumulative reward. ReinboT achieves a deeper understanding of the data quality distribution by predicting dense returns that capture the nuances of manipulation tasks. The dense return prediction capability enables the robot to generate more robust decision-making actions, oriented towards maximizing future benefits. Extensive experiments show that ReinboT achieves state-of-the-art performance on the CALVIN mixed-quality dataset and exhibits superior few-shot learning and out-of-distribution generalization capabilities in real-world tasks.

Autonomous racing presents unique challenges due to its non-linear dynamics, the high speed involved, and the critical need for real-time decision-making under dynamic and unpredictable conditions. Most traditional Reinforcement Learning (RL) approaches rely on extensive simulation-based pre-training, which faces crucial challenges in transfer effectively to real-world environments. This paper introduces a robust on-board RL framework for autonomous racing, designed to eliminate the dependency on simulation-based pre-training enabling direct real-world adaptation. The proposed system introduces a refined Soft Actor-Critic (SAC) algorithm, leveraging a residual RL structure to enhance classical controllers in real-time by integrating multi-step Temporal-Difference (TD) learning, an asynchronous training pipeline, and Heuristic Delayed Reward Adjustment (HDRA) to improve sample efficiency and training stability. The framework is validated through extensive experiments on the F1TENTH racing platform, where the residual RL controller consistently outperforms the baseline controllers and achieves up to an 11.5 % reduction in lap times compared to the State-of-the-Art (SotA) with only 20 min of training. Additionally, an End-to-End (E2E) RL controller trained without a baseline controller surpasses the previous best results with sustained on-track learning.

We introduce INTELLECT-2, the first globally distributed reinforcement learning (RL) training run of a 32 billion parameter language model. Unlike traditional centralized training efforts, INTELLECT-2 trains a reasoning model using fully asynchronous RL across a dynamic, heterogeneous swarm of permissionless compute contributors. To enable a training run with this unique infrastructure, we built various components from scratch: we introduce PRIME-RL, our training framework purpose-built for distributed asynchronous reinforcement learning, based on top of novel components such as TOPLOC, which verifies rollouts from untrusted inference workers, and SHARDCAST, which efficiently broadcasts policy weights from training nodes to inference workers. Beyond infrastructure components, we propose modifications to the standard GRPO training recipe and data filtering techniques that were crucial to achieve training stability and ensure that our model successfully learned its training objective, thus improving upon QwQ-32B, the state of the art reasoning model in the 32B parameter range. We open-source INTELLECT-2 along with all of our code and data, hoping to encourage and enable more open research in the field of decentralized training.