Dynamic Residual Safe Reinforcement Learning for Multi-Agent Safety-Critical Scenarios Decision-Making

In multi-agent safety-critical scenarios, traditional autonomous driving frameworks face significant challenges in balancing safety constraints and task performance. These frameworks struggle to quantify dynamic interaction risks in real-time and depend heavily on manual rules, resulting in low computational efficiency and conservative strategies. To address these limitations, we propose a Dynamic Residual Safe Reinforcement Learning (DRS-RL) framework grounded in a safety-enhanced networked Markov decision process. It's the first time that the weak-to-strong theory is introduced into multi-agent decision-making, enabling lightweight dynamic calibration of safety boundaries via a weak-to-strong safety correction paradigm. Based on the multi-agent dynamic conflict zone model, our framework accurately captures spatiotemporal coupling risks among heterogeneous traffic participants and surpasses the static constraints of conventional geometric rules. Moreover, a risk-aware prioritized experience replay mechanism mitigates data distribution bias by mapping risk to sampling probability. Experimental results reveal that the proposed method significantly outperforms traditional RL algorithms in safety, efficiency, and comfort. Specifically, it reduces the collision rate by up to 92.17%, while the safety model accounts for merely 27% of the main model's parameters.

Graph Based Deep Reinforcement Learning Aided by Transformers for Multi-Agent Cooperation

Mission planning for a fleet of cooperative autonomous drones in applications that involve serving distributed target points, such as disaster response, environmental monitoring, and surveillance, is challenging, especially under partial observability, limited communication range, and uncertain environments. Traditional path-planning algorithms struggle in these scenarios, particularly when prior information is not available. To address these challenges, we propose a novel framework that integrates Graph Neural Networks (GNNs), Deep Reinforcement Learning (DRL), and transformer-based mechanisms for enhanced multi-agent coordination and collective task execution. Our approach leverages GNNs to model agent-agent and agent-goal interactions through adaptive graph construction, enabling efficient information aggregation and decision-making under constrained communication. A transformer-based message-passing mechanism, augmented with edge-feature-enhanced attention, captures complex interaction patterns, while a Double Deep Q-Network (Double DQN) with prioritized experience replay optimizes agent policies in partially observable environments. This integration is carefully designed to address specific requirements of multi-agent navigation, such as scalability, adaptability, and efficient task execution. Experimental results demonstrate superior performance, with 90% service provisioning and 100% grid coverage (node discovery), while reducing the average steps per episode to 200, compared to 600 for benchmark methods such as particle swarm optimization (PSO), greedy algorithms and DQN.

Grasping Deformable Objects via Reinforcement Learning with Cross-Modal Attention to Visuo-Tactile Inputs

We consider the problem of grasping deformable objects with soft shells using a robotic gripper. Such objects have a center-of-mass that changes dynamically and are fragile so prone to burst. Thus, it is difficult for robots to generate appropriate control inputs not to drop or break the object while performing manipulation tasks. Multi-modal sensing data could help understand the grasping state through global information (e.g., shapes, pose) from visual data and local information around the contact (e.g., pressure) from tactile data. Although they have complementary information that can be beneficial to use together, fusing them is difficult owing to their different properties. We propose a method based on deep reinforcement learning (DRL) that generates control inputs of a simple gripper from visuo-tactile sensing information. Our method employs a cross-modal attention module in the encoder network and trains it in a self-supervised manner using the loss function of the RL agent. With the multi-modal fusion, the proposed method can learn the representation for the DRL agent from the visuo-tactile sensory data. The experimental result shows that cross-modal attention is effective to outperform other early and late data fusion methods across different environments including unseen robot motions and objects.

Human-like compositional learning of visually-grounded concepts using synthetic environments

The compositional structure of language enables humans to decompose complex phrases and map them to novel visual concepts, showcasing flexible intelligence. While several algorithms exhibit compositionality, they fail to elucidate how humans learn to compose concept classes and ground visual cues through trial and error. To investigate this multi-modal learning challenge, we designed a 3D synthetic environment in which an agent learns, via reinforcement, to navigate to a target specified by a natural language instruction. These instructions comprise nouns, attributes, and critically, determiners, prepositions, or both. The vast array of word combinations heightens the compositional complexity of the visual grounding task, as navigating to a blue cube above red spheres is not rewarded when the instruction specifies navigating to "some blue cubes below the red sphere". We first demonstrate that reinforcement learning agents can ground determiner concepts to visual targets but struggle with more complex prepositional concepts. Second, we show that curriculum learning, a strategy humans employ, enhances concept learning efficiency, reducing the required training episodes by 15% in determiner environments and enabling agents to easily learn prepositional concepts. Finally, we establish that agents trained on determiner or prepositional concepts can decompose held-out test instructions and rapidly adapt their navigation policies to unseen visual object combinations. Leveraging synthetic environments, our findings demonstrate that multi-modal reinforcement learning agents can achieve compositional understanding of complex concept classes and highlight the efficacy of human-like learning strategies in improving artificial systems' learning efficiency.

Two-Agent DRL for Power Allocation and IRS Orientation in Dynamic NOMA-based OWC Networks

Intelligent reflecting surfaces (IRSs) technology has been considered a promising solution in visible light communication (VLC) systems due to its potential to overcome the line-of-sight (LoS) blockage issue and enhance coverage. Moreover, integrating IRS with a downlink non-orthogonal multiple access (NOMA) transmission technique for multi-users is a smart solution to achieve a high sum rate and improve system performance. In this paper, a dynamic IRS-assisted NOMA-VLC system is modeled, and an optimization problem is formulated to maximize sum energy efficiency (SEE) and fairness among multiple mobile users under power allocation and IRS mirror orientation constraints. Due to the non-convex nature of the optimization problem and the non-linearity of the constraints, conventional optimization methods are impractical for real-time solutions. Therefore, a two-agent deep reinforcement learning (DRL) algorithm is designed for optimizing power allocation and IRS orientation based on centralized training with decentralized execution to obtain fast and real-time solutions in dynamic environments. The results show the superior performance of the proposed DRL algorithm compared to standard DRL algorithms typically used for resource allocation in wireless communication. The results also show that the proposed DRL algorithm achieves higher performance compared to deployments without IRS and with randomly oriented IRS elements.

Leveraging LLMs as Meta-Judges: A Multi-Agent Framework for Evaluating LLM Judgments

Large language models (LLMs) are being widely applied across various fields, but as tasks become more complex, evaluating their responses is increasingly challenging. Compared to human evaluators, the use of LLMs to support performance evaluation offers a more efficient alternative. However, most studies focus mainly on aligning LLMs' judgments with human preferences, overlooking the existence of biases and mistakes in human judgment. Furthermore, how to select suitable LLM judgments given multiple potential LLM responses remains underexplored. To address these two aforementioned issues, we propose a three-stage meta-judge selection pipeline: 1) developing a comprehensive rubric with GPT-4 and human experts, 2) using three advanced LLM agents to score judgments, and 3) applying a threshold to filter out low-scoring judgments. Compared to methods using a single LLM as both judge and meta-judge, our pipeline introduces multi-agent collaboration and a more comprehensive rubric. Experimental results on the JudgeBench dataset show about 15.55\% improvement compared to raw judgments and about 8.37\% improvement over the single-agent baseline. Our work demonstrates the potential of LLMs as meta-judges and lays the foundation for future research on constructing preference datasets for LLM-as-a-judge reinforcement learning.

Policy-Based Radiative Transfer: Solving the $2$-Level Atom Non-LTE Problem using Soft Actor-Critic Reinforcement Learning

We present a novel reinforcement learning (RL) approach for solving the classical 2-level atom non-LTE radiative transfer problem by framing it as a control task in which an RL agent learns a depth-dependent source function $S(\tau)$ that self-consistently satisfies the equation of statistical equilibrium (SE). The agent's policy is optimized entirely via reward-based interactions with a radiative transfer engine, without explicit knowledge of the ground truth. This method bypasses the need for constructing approximate lambda operators ($\Lambda^*$) common in accelerated iterative schemes. Additionally, it requires no extensive precomputed labeled datasets to extract a supervisory signal, and avoids backpropagating gradients through the complex RT solver itself. Finally, we show through experiment that a simple feedforward neural network trained greedily cannot solve for SE, possibly due to the moving target nature of the problem. Our $\Lambda^*-\text{Free}$ method offers potential advantages for complex scenarios (e.g., atmospheres with enhanced velocity fields, multi-dimensional geometries, or complex microphysics) where $\Lambda^*$ construction or solver differentiability is challenging. Additionally, the agent can be incentivized to find more efficient policies by manipulating the discount factor, leading to a reprioritization of immediate rewards. If demonstrated to generalize past its training data, this RL framework could serve as an alternative or accelerated formalism to achieve SE. To the best of our knowledge, this study represents the first application of reinforcement learning in solar physics that directly solves for a fundamental physical constraint.

Sky-Drive: A Distributed Multi-Agent Simulation Platform for Socially-Aware and Human-AI Collaborative Future Transportation

Recent advances in autonomous system simulation platforms have significantly enhanced the safe and scalable testing of driving policies. However, existing simulators do not yet fully meet the needs of future transportation research, particularly in modeling socially-aware driving agents and enabling effective human-AI collaboration. This paper introduces Sky-Drive, a novel distributed multi-agent simulation platform that addresses these limitations through four key innovations: (a) a distributed architecture for synchronized simulation across multiple terminals; (b) a multi-modal human-in-the-loop framework integrating diverse sensors to collect rich behavioral data; (c) a human-AI collaboration mechanism supporting continuous and adaptive knowledge exchange; and (d) a digital twin (DT) framework for constructing high-fidelity virtual replicas of real-world transportation environments. Sky-Drive supports diverse applications such as autonomous vehicle (AV)-vulnerable road user (VRU) interaction modeling, human-in-the-loop training, socially-aware reinforcement learning, personalized driving policy, and customized scenario generation. Future extensions will incorporate foundation models for context-aware decision support and hardware-in-the-loop (HIL) testing for real-world validation. By bridging scenario generation, data collection, algorithm training, and hardware integration, Sky-Drive has the potential to become a foundational platform for the next generation of socially-aware and human-centered autonomous transportation research. The demo video and code are available at:https://sky-lab-uw.github.io/Sky-Drive-website/

SwitchMT: An Adaptive Context Switching Methodology for Scalable Multi-Task Learning in Intelligent Autonomous Agents

The ability to train intelligent autonomous agents (such as mobile robots) on multiple tasks is crucial for adapting to dynamic real-world environments. However, state-of-the-art reinforcement learning (RL) methods only excel in single-task settings, and still struggle to generalize across multiple tasks due to task interference. Moreover, real-world environments also demand the agents to have data stream processing capabilities. Toward this, a state-of-the-art work employs Spiking Neural Networks (SNNs) to improve multi-task learning by exploiting temporal information in data stream, while enabling lowpower/energy event-based operations. However, it relies on fixed context/task-switching intervals during its training, hence limiting the scalability and effectiveness of multi-task learning. To address these limitations, we propose SwitchMT, a novel adaptive task-switching methodology for RL-based multi-task learning in autonomous agents. Specifically, SwitchMT employs the following key ideas: (1) a Deep Spiking Q-Network with active dendrites and dueling structure, that utilizes task-specific context signals to create specialized sub-networks; and (2) an adaptive task-switching policy that leverages both rewards and internal dynamics of the network parameters. Experimental results demonstrate that SwitchMT achieves superior performance in multi-task learning compared to state-of-the-art methods. It achieves competitive scores in multiple Atari games (i.e., Pong: -8.8, Breakout: 5.6, and Enduro: 355.2) compared to the state-of-the-art, showing its better generalized learning capability. These results highlight the effectiveness of our SwitchMT methodology in addressing task interference while enabling multi-task learning automation through adaptive task switching, thereby paving the way for more efficient generalist agents with scalable multi-task learning capabilities.

Cross-environment Cooperation Enables Zero-shot Multi-agent Coordination

Zero-shot coordination (ZSC), the ability to adapt to a new partner in a cooperative task, is a critical component of human-compatible AI. While prior work has focused on training agents to cooperate on a single task, these specialized models do not generalize to new tasks, even if they are highly similar. Here, we study how reinforcement learning on a distribution of environments with a single partner enables learning general cooperative skills that support ZSC with many new partners on many new problems. We introduce two Jax-based, procedural generators that create billions of solvable coordination challenges. We develop a new paradigm called Cross-Environment Cooperation (CEC), and show that it outperforms competitive baselines quantitatively and qualitatively when collaborating with real people. Our findings suggest that learning to collaborate across many unique scenarios encourages agents to develop general norms, which prove effective for collaboration with different partners. Together, our results suggest a new route toward designing generalist cooperative agents capable of interacting with humans without requiring human data.