Legged robots offer mobility where wheeled platforms fail, such as stairs, rubble, soft substrates, and cluttered indoor-outdoor settings, enabling applications in inspection, search and rescue, agriculture, and planetary exploration [1]. Robust locomotion control is therefore a foundational capability for practical quadrupedal systems, underpinning safe navigation and dependable operation across diverse terrains and disturbances [2]. Deep reinforcement learning (DRL) has emerged as a compelling paradigm for such control because it optimizes closed-loop policies through interaction and can produce adaptive behaviors [3]. A substantial body of prior work has focused on training blind controllers that rely exclusively on proprioceptive inputs such as inertial measurement units (IMUs) and joint feedback [4]. While these blind policies can traverse uneven and unknown terrains through large-scale simulation and domain randomization, they inherently lack foresight: without exteroceptive input, they respond only upon contact and struggle to proactively avoid obstacles or plan foot placement on irregular ground. Vision complements proprioception by providing anticipatory geometric information, enabling early detection of distant obstacles and terrain changes [5]. As a result, cross-modal policies that integrate proprioception with depth imaging have gained prominence, facilitating safer and more efficient locomotion through earlier trajectory adjustments. Most existing cross-modal pipelines adopt multilayer perceptrons (MLPs) for the proprioceptive encoder and for the decision head that fuses proprioception with vision.

Multi-agent systems (MAS) built on multimodal large language models exhibit strong collaboration and performance. However, their growing openness and interaction complexity pose serious risks, notably jailbreak and adversarial attacks. Existing defenses typically rely on external guard modules, such as dedicated safety agents, to handle unsafe behaviors. Unfortunately, this paradigm faces two challenges: (1) standalone agents offer limited protection, and (2) their independence leads to single-point failure-if compromised, system-wide safety collapses. Naively increasing the number of guard agents further raises cost and complexity. To address these challenges, we propose Evo-MARL, a novel multi-agent reinforcement learning (MARL) framework that enables all task agents to jointly acquire defensive capabilities. Rather than relying on external safety modules, Evo-MARL trains each agent to simultaneously perform its primary function and resist adversarial threats, ensuring robustness without increasing system overhead or single-node failure. Furthermore, Evo-MARL integrates evolutionary search with parameter-sharing reinforcement learning to co-evolve attackers and defenders. This adversarial training paradigm internalizes safety mechanisms and continually enhances MAS performance under co-evolving threats. Experiments show that Evo-MARL reduces attack success rates by up to 22% while boosting accuracy by up to 5% on reasoning tasks-demonstrating that safety and utility can be jointly improved.

Multimodal Large Language Models (MLLMs) already achieve state-of-the-art results across a wide range of tasks and modalities. T o push their reasoning ability further, recent studies explore advanced prompting schemes and post-training fine-tuning. Although these techniques improve logical accuracy, they frequently leave the models' outputs burdened with pronounced social biases. Clarifying how reasoning gains interact with bias mitigation--and whether the two objectives inherently trade off--therefore remains an open and pressing research problem. Our study begins by benchmarking three bias-mitigation strategies--supervised fine-tuning (SFT), knowledge distillation (KD), and rule-based reinforcement learning (RL)--under identical conditions, establishing their baseline strengths and weaknesses. Building on these results, we vary the proportion of debias-focused and reasoning-centric samples within each paradigm to chart the reasoning-versus-bias trade-off. Our sweeps reveal a consistent sweet spot: a roughly 1:4 mix trained with reinforcement learning cuts stereotype scores by 10% while retaining 88% of the model's original reasoning accuracy, offering concrete guidance for balancing fairness and capability in MLLMs.

Behavior Cloning (BC) on curated (or filtered) data is the predominant paradigm for supervised fine-tuning (SFT) of large language models; as well as for imitation learning of control policies. Here, we draw on a connection between this successful strategy and the theory and practice of finding optimal policies via Reinforcement Learning (RL). Building on existing literature, we clarify that SFT can be understood as maximizing a lower bound on the RL objective in a sparse reward setting. Giving support to its often observed good performance. From this viewpoint, we realize that a small modification to SFT leads to an importance weighted variant that behaves closer to training with RL as it: i) optimizes a tighter bound to the RL objective and, ii) can improve performance compared to SFT on curated data. We refer to this variant as importance weighted supervised fine-tuning (iw-SFT). We show that it is easy to implement and can be further generalized to training with quality scored data. The resulting SFT variants are competitive with more advanced RL algorithms for large language models and for training policies in continuous control tasks. For example achieving 66.7% on the AIME 2024 dataset.

Accurate bandwidth estimation (BWE) is critical for real-time communication (RTC) systems. Traditional heuristic approaches offer limited adaptability under dynamic networks, while online reinforcement learning (RL) suffers from high exploration costs and potential service disruptions. Offline RL, which leverages high-quality data collected from real-world environments, offers a promising alternative. However, challenges such as out-of-distribution (OOD) actions, policy extraction from behaviorally diverse datasets, and reliable deployment in production systems remain unsolved. We propose RBWE, a robust bandwidth estimation framework based on offline RL that integrates Q-ensemble (an ensemble of Q-functions) with a Gaussian mixture policy to mitigate OOD risks and enhance policy learning. A fallback mechanism ensures deployment stability by switching to heuristic methods under high uncertainty. Experimental results show that RBWE reduces overestimation errors by 18% and improves the 10th percentile Quality of Experience (QoE) by 18.6%, demonstrating its practical effectiveness in real-world RTC applications. The implementation is publicly available at https://github.com/jiu2021/RBWE_offline.

Benchmarks play a crucial role in the development and analysis of reinforcement learning (RL) algorithms, with environment availability strongly impacting research. One particularly underexplored intersection is continual learning (CL) in cooperative multi-agent settings. To remedy this, we introduce MEAL (Multi-agent Environments for Adaptive Learning), the first benchmark tailored for continual multi-agent reinforcement learning (CMARL). Existing CL benchmarks run environments on the CPU, leading to computational bottlenecks and limiting the length of task sequences. MEAL leverages JAX for GPU acceleration, enabling continual learning across sequences of 100 tasks on a standard desktop PC in a few hours. We show that naively combining popular CL and MARL methods yields strong performance on simple environments, but fails to scale to more complex settings requiring sustained coordination and adaptation. Our ablation study identifies architectural and algorithmic features critical for CMARL on MEAL.

A universal controller for any robot morphology would greatly improve computational and data efficiency. By utilizing contextual information about the properties of individual robots and exploiting their modular structure in the architecture of deep reinforcement learning agents, steps have been made towards multi-robot control. Generalization to new, unseen robots, however, remains a challenge. In this paper we hypothesize that the relevant contextual information is partially observable, but that it can be inferred through interactions for better generalization to contexts that are not seen during training. To this extent, we implement a modular recurrent architecture and evaluate its generalization performance on a large set of MuJoCo robots. The results show a substantial improved performance on robots with unseen dynamics, kinematics, and topologies, in four different environments.

Climate control is crucial for greenhouse production as it directly affects crop growth and resource use. Reinforcement learning (RL) has received increasing attention in this field, but still faces challenges, including limited training efficiency and high reliance on initial learning conditions. Interactive RL, which combines human (grower) input with the RL agent's learning, offers a potential solution to overcome these challenges. However, interactive RL has not yet been applied to greenhouse climate control and may face challenges related to imperfect inputs. Therefore, this paper aims to explore the possibility and performance of applying interactive RL with imperfect inputs into greenhouse climate control, by: (1) developing three representative interactive RL algorithms tailored for greenhouse climate control (reward shaping, policy shaping and control sharing); (2) analyzing how input characteristics are often contradicting, and how the trade-offs between them make grower's inputs difficult to perfect; (3) proposing a neural network-based approach to enhance the robustness of interactive RL agents under limited input availability; (4) conducting a comprehensive evaluation of the three interactive RL algorithms with imperfect inputs in a simulated greenhouse environment. The demonstration shows that interactive RL incorporating imperfect grower inputs has the potential to improve the performance of the RL agent. RL algorithms that influence action selection, such as policy shaping and control sharing, perform better when dealing with imperfect inputs, achieving 8.4% and 6.8% improvement in profit, respectively. In contrast, reward shaping, an algorithm that manipulates the reward function, is sensitive to imperfect inputs and leads to a 9.4% decrease in profit. This highlights the importance of selecting an appropriate mechanism when incorporating imperfect inputs.

Continual learning in robotics seeks systems that can constantly adapt to changing environments and tasks, mirroring human adaptability. A key challenge is refining dynamics models, essential for planning and control, while addressing issues such as safe adaptation, catastrophic forgetting, outlier management, data efficiency, and balancing exploration with exploitation -- all within task and onboard resource constraints. Towards this goal, we introduce a generative framework leveraging flow matching for online robot dynamics model alignment. Rather than executing actions based on a misaligned model, our approach refines planned actions to better match with those the robot would take if its model was well aligned. We find that by transforming the actions themselves rather than exploring with a misaligned model -- as is traditionally done -- the robot collects informative data more efficiently, thereby accelerating learning. Moreover, we validate that the method can handle an evolving and possibly imperfect model while reducing, if desired, the dependency on replay buffers or legacy model snapshots. We validate our approach using two platforms: an unmanned ground vehicle and a quadrotor. The results highlight the method's adaptability and efficiency, with a record 34.2\% higher task success rate, demonstrating its potential towards enabling continual robot learning. Code: https://github.com/AlejandroMllo/action_flow_matching.

Recent advancements in image customization exhibit a wide range of application prospects due to stronger customization capabilities. However, since we humans are more sensitive to faces, a significant challenge remains in preserving consistent identity while avoiding identity confusion with multi-reference images, limiting the identity scalability of customization models. To address this, we present UMO, a Unified Multi-identity Optimization framework, designed to maintain high-fidelity identity preservation and alleviate identity confusion with scalability. With "multi-to-multi matching" paradigm, UMO reformulates multi-identity generation as a global assignment optimization problem and unleashes multi-identity consistency for existing image customization methods generally through reinforcement learning on diffusion models. To facilitate the training of UMO, we develop a scalable customization dataset with multi-reference images, consisting of both synthesised and real parts. Additionally, we propose a new metric to measure identity confusion. Extensive experiments demonstrate that UMO not only improves identity consistency significantly, but also reduces identity confusion on several image customization methods, setting a new state-of-the-art among open-source methods along the dimension of identity preserving. Code and model: https://github.com/bytedance/UMO