Multi-Agent Path Finding (MAPF) requires computing collision-free paths for multiple agents in shared environment. Most MAPF planners assume that each agent reaches a specific location at a specific timestep, but this is infeasible to directly follow on real systems where delays often occur. To address collisions caused by agents deviating due to delays, the Temporal Plan Graph (TPG) was proposed, which converts a MAPF time dependent solution into a time independent set of inter-agent dependencies. Recently, a Bidirectional TPG (BTPG) was proposed which relaxed some dependencies into ``bidirectional pairs" and improved efficiency of agents executing their MAPF solution with delays. Our work improves upon this prior work by designing an algorithm, BPTG-max, that finds more bidirectional pairs. Our main theoretical contribution is in designing the BTPG-max algorithm is locally optimal, i.e. which constructs a BTPG where no additional bidirectional pairs can be added. We also show how in practice BTPG-max leads to BTPGs with significantly more bidirectional edges, superior anytime behavior, and improves robustness to delays.

Order dispatch systems play a vital role in ride-hailing services, which directly influence operator revenue, driver profit, and passenger experience. Most existing work focuses on improving system efficiency in terms of operator revenue, which may cause a bad experience for both passengers and drivers. Hence, in this work, we aim to design a human-centered ride-hailing system by considering both passenger fairness and driver preference without compromising the overall system efficiency. However, it is nontrivial to achieve this target due to the potential conflicts between passenger fairness and driver preference since optimizing one may sacrifice the other. To address this challenge, we design HCRide, a H uman-C entered Ride-hailing system based on a novel multi-agent reinforcement learning algorithm called H armonization-oriented A ctor-Bi -C ritic (Habic), which includes three major components (i.e., a multi-agent competition mechanism, a dynamic Actor network, and a Bi-Critic network) to optimize system efficiency and passenger fairness with driver preference consideration. We extensively evaluate our HCRide using two real-world ride-hailing datasets from Shenzhen and New Y ork City. Experimental results show our HCRide effectively improves system efficiency by 2.02%, fairness by 5.39%, and driver preference by 10.21% compared to state-of-the-art baselines.

We study an online fair division problem where a fixed number of goods arrive sequentially and must be allocated to a given set of agents. Once a good arrives, its true value for each agent is revealed, and it has to be immediately and irrevocably allocated to some agent. The ultimate goal is to ensure envy-freeness up to any good (EFX) after all goods have been allocated. Unfortunately, as we show, approximate EFX allocations are unattainable in general, even under restrictive assumptions on the valuation functions. To address this, we follow a recent and fruitful trend of augmenting algorithms with predictions. Specifically, we assume access to a prediction vector estimating the agents' true valuations -- e.g., generated by a machine learning model trained on past data. Predictions may be unreliable, and we measure their error using the total variation distance from the true valuations, that is, the percentage of predicted value-mass that disagrees with the true values. Focusing on the natural class of additive valuations, we prove impossibility results even on approximate EFX allocations for algorithms that either ignore predictions or rely solely on them. We then turn to algorithms that use both the predictions and the true values and show strong lower bounds on the prediction accuracy that is required by any algorithm to compute an approximate EFX. These negative results persist even under identical valuations, contrary to the offline setting where exact EFX allocations always exist without the necessity of predictions. We then present an algorithm for two agents with identical valuations that uses effectively the predictions and the true values. The algorithm approximates EFX, with its guarantees improving as the accuracy of the predictions increases.

We present GeoFlow, a method that automatically generates agentic workflows for geospatial tasks. Unlike prior work that focuses on reasoning decomposition and leaves API selection implicit, our method provides each agent with detailed tool-calling objectives to guide geospatial API invocation at runtime. GeoFlow increases agentic success by 6.8% and reduces token usage by up to fourfold across major LLM families compared to state-of-the-art approaches.

Most prior works on communication in multi-agent reinforcement learning have focused on emergent communication, which often results in inefficient and non-interpretable systems. Inspired by the role of language in natural intelligence, we investigate how grounding agents in a human-defined language can improve the learning and coordination of embodied agents. We propose a framework in which agents are trained not only to act but also to produce and interpret natural language descriptions of their observations. This language-augmented learning serves a dual role: enabling efficient and interpretable communication between agents, and guiding representation learning. We demonstrate that language-augmented agents outperform emergent communication baselines across various tasks. Our analysis reveals that language grounding leads to more informative internal representations, better generalization to new partners, and improved capability for human-agent interaction. These findings demonstrate the effectiveness of integrating structured language into multi-agent learning and open avenues for more interpretable and capable multi-agent systems.

We have seen remarkable progress in large language models (LLMs) empowered multi-agent systems solving complex tasks necessitating cooperation among experts with diverse skills. However, optimizing LLM-based multi-agent systems remains challenging. In this work, we perform an empirical case study on group optimization of role-based multi-agent systems utilizing natural language feedback for challenging software development tasks under various evaluation dimensions. We propose a two-step agent prompts optimization pipeline: identifying underperforming agents with their failure explanations utilizing textual feedback and then optimizing system prompts of identified agents utilizing failure explanations. We then study the impact of various optimization settings on system performance with two comparison groups: online against offline optimization and individual against group optimization. For group optimization, we study two prompting strategies: one-pass and multi-pass prompting optimizations. Overall, we demonstrate the effectiveness of our optimization method for role-based multi-agent systems tackling software development tasks evaluated on diverse evaluation dimensions, and we investigate the impact of diverse optimization settings on group behaviors of the multi-agent systems to provide practical insights for future development.

This work presents a novel approach for \textit{bearing rigidity} analysis and control in multi-robot networks with sensing constraints and dynamic topology. By decomposing the system's framework into \textit{subframeworks}, we express bearing rigidity -- a global property -- as a set of \textit{local} properties, with rigidity eigenvalues serving as natural \textit{local rigidity measures}. We propose a decentralized gradient-based controller to execute mission-specific commands using only bearing measurements. The controller preserves bearing rigidity by keeping the rigidity eigenvalues above a threshold, using only information exchanged within subframeworks. Simulations evaluate the scheme's effectiveness, underscoring its scalability and practicality.

We propose a hybrid architecture that integrates decision tree-based symbolic reasoning with the generative capabilities of large language models (LLMs) within a coordinated multi-agent framework. Unlike prior approaches that loosely couple symbolic and neural modules, our design embeds decision trees and random forests as callable oracles within a unified reasoning system. Tree-based modules enable interpretable rule inference and causal logic, while LLM agents handle abductive reasoning, generalization, and interactive planning. A central orchestrator maintains belief state consistency and mediates communication across agents and external tools, enabling reasoning over both structured and unstructured inputs. The system achieves strong performance on reasoning benchmarks. On \textit{ProofWriter}, it improves entailment consistency by +7.2\% through logic-grounded tree validation. On GSM8k, it achieves +5.3\% accuracy gains in multistep mathematical problems via symbolic augmentation. On \textit{ARC}, it boosts abstraction accuracy by +6.0\% through integration of symbolic oracles. Applications in clinical decision support and scientific discovery show how the system encodes domain rules symbolically while leveraging LLMs for contextual inference and hypothesis generation. This architecture offers a robust, interpretable, and extensible solution for general-purpose neuro-symbolic reasoning.

Generative AI is no longer a peripheral tool in higher education. It is rapidly evolving into a general-purpose infrastructure that reshapes how knowledge is generated, mediated, and validated. This paper presents findings from a controlled experiment evaluating a Socratic AI Tutor, a large language model designed to scaffold student research question development through structured dialogue grounded in constructivist theory. Conducted with 65 pre-service teacher students in Germany, the study compares interaction with the Socratic Tutor to engagement with an uninstructed AI chatbot. Students using the Socratic Tutor reported significantly greater support for critical, independent, and reflective thinking, suggesting that dialogic AI can stimulate metacognitive engagement and challenging recent narratives of de-skilling due to generative AI usage. These findings serve as a proof of concept for a broader pedagogical shift: the use of multi-agent systems (MAS) composed of specialised AI agents. To conceptualise this, we introduce the notion of orchestrated MAS, modular, pedagogically aligned agent constellations, curated by educators, that support diverse learning trajectories through differentiated roles and coordinated interaction. To anchor this shift, we propose an adapted offer-and-use model, in which students appropriate instructional offers from these agents. Beyond technical feasibility, we examine system-level implications for higher education institutions and students, including funding necessities, changes to faculty roles, curriculars, competencies and assessment practices. We conclude with a comparative cost-effectiveness analysis highlighting the scalability of such systems. In sum, this study contributes both empirical evidence and a conceptual roadmap for hybrid learning ecosystems that embed human-AI co-agency and pedagogical alignment.

Jailbreak attacks against multimodal large language Models (MLLMs) are a significant research focus. Current research predominantly focuses on maximizing attack success rate (ASR), often overlooking whether the generated responses actually fulfill the attacker's malicious intent. This oversight frequently leads to low-quality outputs that bypass safety filters but lack substantial harmful content. To address this gap, we propose JPS, \underline{J}ailbreak MLLMs with collaborative visual \underline{P}erturbation and textual \underline{S}teering, which achieves jailbreaks via corporation of visual image and textually steering prompt. Specifically, JPS utilizes target-guided adversarial image perturbations for effective safety bypass, complemented by "steering prompt" optimized via a multi-agent system to specifically guide LLM responses fulfilling the attackers' intent. These visual and textual components undergo iterative co-optimization for enhanced performance. To evaluate the quality of attack outcomes, we propose the Malicious Intent Fulfillment Rate (MIFR) metric, assessed using a Reasoning-LLM-based evaluator. Our experiments show JPS sets a new state-of-the-art in both ASR and MIFR across various MLLMs and benchmarks, with analyses confirming its efficacy. Codes are available at \href{https://github.com/thu-coai/JPS}{https://github.com/thu-coai/JPS}. \color{warningcolor}{Warning: This paper contains potentially sensitive contents.}