game ai programmers

Abstract: Experience can help embodied agents turn high-level plans into grounded actions, but transferring it across tasks remains challenging: previously effective strategies may be inapplicable to the current state or fail to produce their intended outcomes. We present EmbodiExp, a framework for constructing transferable experiences and composing them online during embodied task execution. Our key observation is that tasks often share subgoals whose completion can be assessed independently of overall task success. We therefore introduce an outcome-grounded subgoal experience that specifies a subgoal, applicability conditions, an adaptable policy, and a verifiable outcome. During execution, applicable experiences guide action generation, while verified outcomes can establish the conditions required by subsequent experiences, enabling online composition. Subgoal-level outcome attribution further enables failure utilization: failed trajectories can still contribute verified local successes and condition-specific guidance on actions to avoid. Experiments on multiple embodied benchmarks under a fold-based protocol with frozen experience banks demonstrate the effectiveness of EmbodiExp.

Abstract: Latent multi-agent systems reduce inference cost by keeping intermediate reasoning and inter-agent communication in continuous representations.\ Reusable skills, however, are commonly supplied as text or stored as layer-wise Key-Value states or skill-specific model parameters, introducing repeated processing or substantial per-skill storage.\ We introduce \textbf{SkillMem}, which represents each reusable skill as a small external latent memory and makes it directly available during latent multi-agent reasoning.\ A role-conditioned cross-attention reader allows the same memory to guide different agents according to their roles and current hidden states.\ We train the skill memories and shared reader through behavioral distillation, matching the change in token predictions induced by a textual skill while keeping the backbone frozen.\ For our experiments, we obtain textual skills from paired successful and failed multi-agent trajectories, although existing skills can also be compiled directly.\ Across six benchmarks and Qwen3 models ranging from 4B to 14B parameters, SkillMem provides strong gains for smaller backbones while preserving efficient latent multi-agent inference across model scales.\ We use the same $8\times256$ memory configuration across all evaluated model scales, requiring 9.6\,KB of skill-specific storage per skill, including its retrieval embedding.\ For a library of 100 skills, SkillMem reduces total storage by more than $38\times$ compared with alternative skill representations and achieves up to a $4.23\times$ end-to-end inference speedup.\ These results show that compact procedural memories can be reused without sacrificing the efficiency of latent multi-agent reasoning.

Abstract: Tool-assisted planning success can reflect external search and verification, leaving a model's unaided planning ability unresolved. We isolate innate planning through complete plans generated in one response without tools or execution feedback. We propose a verifiable framework that decomposes success-weighted plan efficiency as $\mathrm{SPE}=R\times Q\times E$: execution reliability, planning coordination among legal plans, and action economy among completed plans. Trajectory replay and certified optima verify these factors; the joint score includes unsuccessful attempts. We evaluate 8,400 full-plan outputs from seven model conditions on $4\times4$ sliding puzzles, controlling exact horizon and obstacle count in the primary cohort. Models that complete every trial still become less economical at longer horizons, while higher efficiency among completed plans can coexist with poor goal attainment. SPE separates the overall capability hierarchy, while its factors distinguish coordination failure from inefficient completion. The framework makes long-horizon limitations in unaided planning quantitatively verifiable.

Abstract: Hierarchical task knowledge enables efficient problem solving by allowing previously acquired skills to be reused across tasks and problem instances. In this paper, we investigate curriculum-guided acquisition of hierarchical task knowledge in an HTN learning framework. We compare three approaches: online learning during problem solving, human-designed curricula, and an automated curriculum generation procedure that derives learning experiences from verified LLM-generated solution traces. The proposed procedure solves small training problems with an LLM, extracts task accomplishment information, and transforms it into ordered sequences of state-task learning problems for HTN method acquisition. We evaluate the three approaches across multiple planning domains. The results demonstrate that automatically generated curricula recover much of the benefit of expert-designed curricula while substantially reducing the token cost of online hierarchical knowledge acquisition.

Abstract: Multi-agent systems rely on communication to complete tasks, and latent communication passes continuous representations to reduce text generation and re-encoding. However, multi-agent systems must reconfigure agent models, whereas existing latent-communication methods are built for fixed model combinations, so their interfaces may fail after replacement. Our pilot study confirms this interface failure and identifies the requirement for interface reuse after model replacement. We present LatBridge, which connects agent models through a shared Bridge and separate interfaces under a shared latent-communication protocol. After agent model replacement, LatBridge reuses unchanged interfaces, selects matching low-rank interfaces, and combines them through the Bridge, training only missing ones with all models frozen. The Bridge processes source representations in the basic mode and constructs target input without autoregressive generation in the accelerated mode. Experiments support effective communication after model replacement, with mean basic-mode accuracies of 14.8% on MATH and 27.3% on ALFWorld next-action prediction across eight small-model configurations. Analyses examine the factors supporting task accuracy and communication efficiency.

Abstract: Reinforcement learning with verifiable rewards has become a scalable post-training paradigm for large language models. However, training agents over long-horizon interactions presents two challenges: using only terminal rewards makes turn-level credit assignment sparse and high-variance, while sequential mini-batch updates on a fixed rollout batch introduce policy staleness as the current policy drifts from the behavior policy. Existing approaches obtain denser credit via learned critics, process reward models, or on-policy distillation, but they rely on a value network, step-level annotations, or a stronger external teacher that increase training complexity. Meanwhile, importance sampling and clipping mitigate policy drift, but they do not specify a finite target that stops sequential updates from overshooting. In this work, we introduce TreeOPSD, a critic-free, tree-aware posterior self-distillation framework for long-horizon agentic reinforcement learning. Treating each interaction turn as a tree node, TreeOPSD branches alternative continuations from shared prefixes and recursively backs up terminal rewards into implicit turn-level credit. It then maps each credit to a closed-form, bounded teacher distribution parameterized by the likelihood ratio relative to the frozen behavior policy. Minimizing the divergence between this teacher and the student policy recovers the policy-gradient direction as the tangent at the behavior policy while bounding updates toward a finite target across sequential mini-batches. Empirically, TreeOPSD improves agentic capabilities across several competitive benchmarks and model scales, especially for long-horizon tasks, e.g., raising the average accuracy of Qwen3-30B-A3B-Thinking-2507 by 4.97 points over GRPO on the Deep Research setting.

Abstract: Generative visual verifiers play a crucial role in advancing multimodal intelligence. To enhance the robustness of visual verifiers in complex multimodal scenarios involving compositional requirements, we move beyond conventional binary verifiers that produce only True/False discrete judgments and introduce a verifier that automatically decomposes complex prompts or environmental rules into fine-grained components and assigns pointwise scores to provide a continuous and more informative critique signal. Our experiments further demonstrate that these structured and interpretable outputs improve the verifier's generalization to out-of-domain multimodal critic tasks. To train this verifier, we introduce a Decoupled RL training strategy that jointly optimizes sub-question-level binary judgment and global alignment scoring, using precise local feedback to develop fine-grained verification capabilities that provide a foundation for reliable continuous scoring at the global level. Under a general reinforcement learning framework, we first empirically demonstrate that the proposed pointwise graded verifier significantly outperforms binary verifiers in traditional text-to-image scenarios. We then extend our investigation to rule-governed generative world modeling where generative models are leveraged for world simulation and world reasoning under explicit rules. From a theoretical standpoint, we further show that a pairwise verification paradigm yields more accurate critiques than the pointwise formulation when ground-truth images are available, with empirical results on diverse structured world-modeling tasks such as Maze and Sudoku further validating the effectiveness of our approach. Overall, our findings suggest a key insight: the transition from generative models to structured world-modeling agents critically hinges on the availability of accurate pairwise graded visual verifiers.

Abstract: World models aim to learn compact representations of an environment from observations, enabling agents to predict future states and plan their behaviour. However, most existing approaches learn black-box representations, with limited ability to incorporate prior knowledge about the world. We introduce Concept-Grounded World Models (CGWMs), a neurosymbolic world model design that instead learns interpretable concept representations. We learn these concepts from observations using explicit and complete knowledge, expressed in temporal logic, that specifies the valid world states and the temporal relations governing their dynamics. Furthermore, we directly use this knowledge by formulating planning as differentiable reasoning, optimising the agent's actions by gradient ascent to satisfy a goal. We experimentally compare CGWM with a purely neural baseline and a neurosymbolic baseline on two game environments. We find that CGWMs ground concepts much more accurately, even over long horizons, resulting in significantly more robust planning both in and out of distribution.

Abstract: Theory of mind (ToM), the ability to reason about others' mental states, underlies effective communication and coordination. Its higher-order form enables reasoning over nested perspectives in strategic interactions. However, existing evaluations often equate belief-nesting depth with reasoning order, allowing nominally higher-order questions to admit lower-order shortcuts. We formalize effective order through the distinct events supporting successive belief prefixes, revealing widespread order inflation across existing benchmarks. We introduce HopToM, with certified full-order questions and controls for lower-order alternatives. Under direct prompting, mean accuracy across four frontier LLMs falls from 73.5% at first order to 20.0% at fourth order, and human participants also struggle. Causal interventions link a recurring class of errors to evidence for shallower belief prefixes. We introduce HopTrace, a plug-and-play inference-time method that encodes each interaction once into perspective records and resolves nested queries deterministically. On certified narrative questions, it improves fourth- and fifth-order accuracy by 22.0 and 38.5 percentage points over strong prompting baselines, while using 20-39 times fewer tokens than multi-call methods. These results sharpen how higher-order ToM is evaluated and offer a practical way to improve perspective reasoning in interactive agents.

Abstract: Relational learning typically predicts from a database state, but decision making asks a different question: how will that state change under an action? In operational systems, the answer depends on system-specific dynamics that may not be visible in the database itself. We view each such system as a relational "world" and ask whether a model can infer how it behaves from observed executions. We introduce TabWM (Tabular World Model) and WorldRel, a generator of executable relational worlds for pretraining. Given executions from an unseen world, TabWM uses them as context to predict the sparse structured edits caused by a new action without parameter updates. Across 512 held-out worlds, complete-edit F1 rises from 0.016 with no observed executions to 0.685 with 64, where a correct edit must match its operation, relational location, and value. Because these predictions are explicit relational edits, they can be applied recursively to imagine candidate futures and compare actions before they are taken. Broader search over these predicted futures improves the realized return of selected action sequences. Together, these results suggest that relational dynamics can be learned in context, broadening relational learning from modeling system state to modeling how that state changes under action.

Abstract: In multi-agent reinforcement learning, exposing a cooperator to partners with diverse strategies during training is key to better collaboration at test time. However, existing methods typically rely on trajectory-level proxies to measure diversity, which may label two distinct trajectories of the same strategy as diverse. Instead, our key insight is to leverage LLMs' ability to describe behaviorally distinct agents in environments, similar to how humans can recognize high-level strategies. To this end, we introduce PopGen, a framework that synthesizes semantically diverse agents from scratch using LLMs. PopGen first generates a diverse set of textual descriptions of agent behaviors, then translates each behavior into a code policy that can be executed within the environment. We show that cooperators trained against PopGen populations achieve robust zero-shot cooperation with unseen partners, outperforming population-play baselines across four collaborative environments. We further show that semantic diversity scales with PopGen population size, improving the cooperator's worst-case performance, and we identify design choices that preserve population quality while maximizing diversity. Our code is available at https://github.com/labicon/PopGen.

Abstract: Beliefs in a Partially Observable Markov Decision Process (POMDP) are posterior distributions over the hidden state given the agent's history of actions and observations. They summarize everything the history says about the state, so planning in a POMDP is planning in a Markov decision process (MDP) over beliefs. Belief MDPs have continuous state spaces even for finite POMDPs, and exact planning in them is intractable in general. Planners must therefore approximate beliefs and estimate values. Monte Carlo planners do both inside a search tree, with an approximate belief at every node and a value estimate at every leaf. Approximations usually consist of sets of sampled states, with leaf values coming from rollouts, hand-designed bounds, or learned networks. Policies and values computed from these approximations are limited by their quality, and prior methods with learned networks rely on hand-chosen summaries of the sampled states. A belief representation should be computable from the parent node's representation through the POMDP's dynamics, let the search sample from it, and be learned so that it provides the information that policies and value functions need. We present Belief Embedding Tree Search (BETS), a Monte Carlo planner that meets these requirements by representing each node's belief with a generative distribution embedding (GDE) while updating it through the POMDP dynamics. BETS trains its belief model alongside its policy and value networks through experience, with the realized hidden state of the POMDP as the belief model's training target. We evaluate BETS on six configurations of RockSample, LightDark, and PocMan against BetaZero and three planners without learned components. BETS is competitive or leading on all six, while every baseline falls far behind on at least one. Where beliefs are highest-dimensional, it outperforms them by a large margin, even without search.

Abstract: In-context learning (ICL) allows large language models to use demonstrations without updating their parameters. For interactive agents, demonstrations are trajectories of reasoning, actions, and observations that illustrate how to complete a task. Reusing demonstrations in new situations is challenging: directly inputting full trajectories consumes limited prompt space and constrains the overall effectiveness, while natural-language summaries require the agent to translate general guidance into specific action choices. Therefore, we study whether expert action choices can directly supervise a reusable scoring rule. We propose CB-Agent, which extends concept-based ICL to sequential action selection. We extract environment-action decisions from expert trajectories and fit a concept vector by maximum likelihood to model the conditional distribution over candidate actions given the context. This model combines the concept with frozen language-model features and priors. During ReAct interaction, the agent uses this distribution to select environment actions without including demonstrations in the action-scoring context. Across TextWorld, ALFWorld, and ScienceWorld, CB-Agent achieves the highest cross-benchmark average among the evaluated methods at all three tested model sizes and ranks first on TextWorld and ScienceWorld at each size. ALFWorld experiments support the learned concept's contribution beyond prior-only scoring and show further gains when combined with Reflexion. Code is available at https://anonymous.4open.science/r/CB-Agent-DFF9/.

Abstract: Self-evolving LLM agents typically improve how they act through accumulated experience, reflection, or skills. Yet even procedurally capable agents can fail when their beliefs about environment dynamics are incomplete, incorrect, or outdated. We study a complementary axis of self-evolution: evolving those beliefs during deployment. Such beliefs can support multiple tasks, but incorrect ones can also propagate across them. We introduce EvoBelief, a self-evolving world model that represents environment knowledge as falsifiable hypotheses. Unexpected outcomes and decision-relevant unknowns trigger hypothesis formation. Later observations and controlled experiments provide independent validation before a rule can guide predictions or plans within its supported scope. Counterexamples narrow or retire rules as the environment changes. The language model remains frozen, and each rule remains linked to its evidence and the decisions it changes. Across four benchmarks spanning static dynamics, hidden drift, and scientific discovery, EvoBelief improves task performance and reuses learned rules across tasks. On ALFWorld without actor-side transition memory, it improves success over the matched Frozen World Model by 23.9% and 31.3% with Claude Haiku 4.5 and GPT-5.4, respectively. On MiniGrid, it revises stale rules and recovers after unannounced dynamics changes. In DiscoveryWorld's Reactor Lab, it discovers and validates quantitative laws, completing 83.3% of Normal-tier and 58.3% of Challenge-tier instances, versus 0% for each evaluated baseline. These results demonstrate self-improvement through testable, revisable environment knowledge rather than model-parameter updates.

Abstract: Planning is increasingly important for long-horizon agents, where successful execution requires coordinating subgoals, tool use, and intermediate outcomes over many steps. Yet assumptions made during planning may be invalidated by the environment, tools may return unexpected results, or actions may fail. Effective agents must therefore not only generate plans, but also revise them. Such revisions often affect only part of a plan, leaving the preceding and subsequent structure intact. Rather than regenerate the entire plan and risk unnecessary changes, repair can regenerate the affected region conditioned on the preserved prefix and suffix. We introduce Plan-and-Patch, a plan-and-act framework in which a diffusion language model (dLLM) generates a structured, program-like plan through parallel unmasking and repairs it by filling in selected regions while keeping the surrounding steps fixed. We compare DreamReasoner-8B and Qwen3-8B as diffusion and autoregressive (AR) planners. On Natural Plan without task-specific training, diffusion (53.7%) achieves nearly twice the plan repair success rate of AR (27.0%). After task-specific training on agentic benchmarks, ALFWorld and TextCraft, the planners achieve similar observed success in plan generation, while diffusion reduces mean plan-generation latency by 39-46% relative to AR. Our results show that Plan-and-Patch provides a framework for faster plan generation and effective plan repair in long-horizon agents.

Abstract: Dual-process agents pair a fast policy with a slow deliberative model. In real-time settings the slow model usually runs continuously; in turn-based agents and robot planners it is invoked on events such as uncertainty or a detected failure. We study a fast learned actor that takes every decision and hands control to a reasoning vision-language model only when a gate opens, while the game keeps running. We use closed-loop Doom and the new open "System One" typed-decision models, served through a common llama.cpp interface. On 900 held-out questions, (i) zero-shot decision models from 0.15B to 9B parameters choose to collect items 1.6-1.8 times more often than chance among their errors, in any option order, although the order changes some models' accuracy; (ii) accuracy, calibration and sensitivity (how well confidence separates right from wrong answers) are distinct: models of similar accuracy differ widely in AUROC, and the confidence of the most sensitive one tracks which kinds of situation it fails, not which answers are wrong; (iii) offline, deferring the least confident 30% of decisions to a reasoning model gains over random deferral in proportion to the actor's AUROC (rank correlation 0.87); with actor and rate chosen on held-out games the gain is +0.13 [0.08, 0.18] with doomLaya's option order and +0.08 [0.02, 0.14] with shuffled options, and reasoning carries about half of it; (iv) in closed loop (33 games, three seeds) no variant reaches the exit. Committing to plans, the reasoner's or a fixed explore rule's, opens more doors and makes an actor that stands still play; with the rule the agent dies more often. Told that some doors need keys, the reasoner takes ordinary doors for locked ones, which the state cannot tell apart; without that knowledge it goes back to collecting. We release code, prompts, data and logs.

Abstract: In partially observable reinforcement learning (RL), a later observation can make stored information obsolete or change what it implies for the next decision. Memory architectures and benchmarks for RL mostly test retention, the ability to keep information unchanged until it is needed. We formalize two further requirements. Rewriting sets the decision-relevant content to a value independent of the old one, and experience fusion transforms the old content by a rule that a later observation specifies. For tasks built from such updates, we count the memory states that a solution needs, and several baselines reach their lowest success rates on compositions that need more states. We introduce ALER (Adaptive Learnable Experience Rewriting), an agent that pairs an LSTM with a slot memory. An independently addressed Gumbel-Softmax write that concentrates its weight on one slot overwrites that slot, and a learned gate fuses the retrieved content with the recurrent state before the policy and value heads. We also introduce Rune-Mazes, three environments in which rune observations invert, cancel, reset, or repeat updates of a hidden cue under vector and pixel observations. Against seven baselines, ALER reaches a success rate of at least $0.82$ in all sixteen Endless T-Maze configurations and at least $0.99$ on all five Rune T-Maze compositions, and it has the highest mean success rate on four-branch Rune Multi-Corridor with an Invert rune. On pixel-based Rune MiniGrid Memory, it has a higher mean success rate than PPO-LSTM in eight of ten configurations. Project page: https://quartz-admirer.github.io/ALER-Adaptive-Learnable-Experience-Rewriting/.

Abstract: Equipping VLM agents with world modeling capabilities has shown strong potential for complex reasoning and long-horizon planning, while reducing the dependence of policy learning on costly real-world interactions. Existing methods mainly rely on prospective simulation to predict the consequences of candidate actions. However, this forward-only paradigm focuses on what will happen next and provides limited constraints for verifying whether an action is causally consistent with the observed state transition, which can lead to plausible-looking but physically incoherent behaviors. In this paper, we challenge the view of world modeling as only prospective prediction and introduce Retrospective World Modeling, a new agent learning paradigm that enables agents to reason backward by estimating the retrospective attribution distribution $P(\hat{a}{t}|s_t, s{t+1})$ for the action that most likely caused a given transition. Based on this capability, we formulate the Self-Consistency Reward (SCR), an intrinsic signal that measures the probabilistic consistency between the policy action and the retrospective explanation. Integrating SCR into reinforcement learning provides dense transition-level feedback and steers agents toward behaviors that are both task-effective and physically grounded. Extensive experiments across diverse agentic tasks show that our method substantially improves policy robustness and generalization over prospective-only world modeling baselines.

Abstract: A message drawn from a small alphabet helps a teammate only if it keeps the distinctions that change that teammate's next decision. We show that scoring messages by action values averaged over the receiver's situation can erase exactly these distinctions, and we propose RAVEN (Receiver-conditioned Action-Value ENcoding), which trains a four-symbol, one-step-delayed channel to preserve each receiver's centered action-value profile within the receiver's own context. The sender never needs to know that context: the receiver decodes every symbol with its private information. We give two estimators of this target. With a teacher, offline RAVEN selects the codebook that exactly minimizes an empirical conditional distortion and distills it into a frozen sender; we bound the resulting codebook-selection error and one-step decision loss. Without a teacher, online RAVEN aligns, inside a QMIX learner, the deployed symbol pathway with a training-only continuous reference that shares its routing. Against five recent communication methods on eight navigation settings, offline RAVEN attains the highest return in seven, and removing receiver conditioning forfeits 83% of its communication gain. Online RAVEN raises predator-prey capture success from 53.2% to 96.0% over the same QMIX backbone without communication, and on SMAC and MPE it attains the best mean normalized score of 14 methods, including methods that exchange kilobit messages. Every RAVEN message costs 2 bits, 12-1,024x fewer than those of NDQ, CACOM and ExpoComm on navigation.

Abstract: On-policy distillation (OPD) trains multi-turn language agents with dense teacher supervision on student-generated responses. However, standard think-then-act rollouts require lengthy reasoning before each short action, delaying environment transitions and experience collection. Generating actions directly reduces this delay but can degrade rollout quality. To address this, we propose ActFirst-OPD, an act-first, reason-later training framework that decouples environment interaction from full-response generation. The student infers and executes actions through reference-conditioned inverse dynamics using its current interaction context and a reference next observation, and switches to autonomous next-action prediction when the resulting transition deviates from the reference trajectory. From the collected interaction contexts, the student asynchronously generates full think-then-act responses for token-level teacher supervision. Experiments across 0.6B-, 1.7B-, and 4B-parameter Qwen3 students show that ActFirst-OPD achieves average wall-clock training speedups of $2.3\times$ on ALFWorld, $1.8\times$ on WebShop, and $4.9\times$ on ScienceWorld over Vanilla OPD. It matches or exceeds all compared OPD baselines in mean task success rate across eight of nine benchmark-model settings. These results demonstrate that reasoning need not block acting during multi-turn agent distillation.

Abstract: Long-horizon agents require context compression to manage growing interaction histories. Compression quality, however, is ultimately determined by downstream execution. Existing prompt-adaptation methods infer compression errors by comparing full-context and compressed trajectories. Such comparisons cannot isolate individual compressions and are confounded by agent stochasticity. We first find that compression degrades reliability before solvability. Using matched counterfactual continuations that compare execution from the same agent state with versus without compression, we further show that severe degradation concentrates at isolated compression events. Motivated by this finding, we propose PAIR (Prompt Adaptation using Interventional Rollouts) for adapting structured compression prompts. PAIR identifies individual compressions that degrade subsequent execution, diagnoses their effects, and revises the relevant sections of a fixed compression template. PAIR achieves the strongest cross-run reliability among compressed methods in every main benchmark-scope combination, consistently exceeding the competing prompt-adaptation baseline. Without modifying the downstream agent, PAIR brings compressed execution close to the no-compression baseline and sometimes numerically exceeds it.

Abstract: World models predict future observations from current experience and actions, yet prediction can depend on observations seen far in the past. Episodic memory preserves past observations for later recall; however, as memory accumulates, it raises a fundamental question: which memories are useful for the current prediction, and which available retrieval cues should be trusted to find them? This is challenging because fixed criteria based on recency, pose overlap, or visual similarity can be unreliable across environments and queries. We propose Future-Aware Recall (FAR), a framework that learns episodic recall from future-aware predictive supervision and adaptive multi-cue scoring. During training, FAR measures predictive utility by the conditional log-likelihood of the realized future given recalled context, approximated by negative diffusion prediction loss, and uses it to train a retriever that remains future-blind at inference. The retriever learns cue-specific relevance and automatically determines which available retrieval cues, such as time, pose, vision, and audio, to trust for each query when selecting memories. Across three complementary settings, FAR outperforms hand-designed recall even with the same retrieval cues, automatically adapts which available cues to trust, and recalls the right history as the world changes. Together, these results establish FAR as a flexible, principled approach to episodic memory access in world models.

Abstract: Despite rapid progress in vision-language-action (VLA) models, existing reasoning paradigms still face a fundamental \emph{state-representation mismatch} in open-loop planning. Given only an initial observation, models must internally simulate action-conditioned state transitions, whereas text-, pixel-, and latent-space reasoning can suffer from lossy spatial compression, error-accumulating visual generation, and bypass of intermediate latent tokens, respectively, undermining reliable long-horizon planning. We propose \textbf{State-Space Visual Reasoning} (SSVR), which decouples static visual context, language constraints, and a recurrent latent state. SSVR encodes the initial image and instruction once, then conditions each action prediction on the latent state and updates it with an action-conditioned GRU. Using Qwen2.5-VL as the backbone, SSVR achieves 99.5/99.6, 96.3/98.0, and 83.9/90.6 EM/PR on FrozenLake, Maze, and MiniBehavior, substantially outperforming prior methods. Extensive experiments support the effectiveness of recurrent state modeling for VLA open-loop planning across input transformations and transfer settings. By reusing static visual-textual context and updating a compact recurrent state, SSVR supports efficient multi-step inference, achieving up to $98.58\times$ faster Maze decoding rollouts than the evaluated baselines with the prefix cache prebuilt.

Abstract: GUI World Models (GUI-WMs) are increasingly used to predict future states for agent planning and simulation, yet most existing formulations condition only on the current GUI observation and action. We identify state aliasing, where the vis- ible interface omits transition-relevant environment state, so identical observable conditions can correspond to different valid futures. To diagnose this failure mode, we introduce StateAliasBench, a diagnostic benchmark that explicitly isolates such ambiguities via strict pairing. We further propose lightweight predictive- state recovery that infers structured state from history and augments otherwise frozen GUI-WMs through a deterministic state interface. Family-specific special- ists provide state recovery across heterogeneous state types, and multi-teacher dis- tillation consolidates them into a single unified estimator. Experiments show that existing GUI-WMs exhibit systematic failures under observation-only condition- ing, while predictive-state augmentation substantially restores state-sensitive pre- diction across evaluated WMs, preserves generative fidelity, and improves down- stream performance of GUI agents on AndroidWorld. These results suggest that reliable GUI world modeling should account not only for what is visible, but also for the hidden transition state that determines what happens next.

Abstract: Model misspecification poses a fundamental challenge in multi-agent reinforcement learning, where transition uncertainty can be amplified by strategic interactions among agents. Distributionally robust Markov games (DRMGs) provide a principled framework for addressing such uncertainty, yet existing methods often rely on restrictive assumptions or scale poorly to large state and joint action spaces. We study online learning in general-sum DRMGs with general function approximation and $φ$-divergence uncertainty sets. We propose RoMEX-$φ$, a model-free framework that integrates equilibrium-based exploration with dual fitted learning. Through a functional dual representation of the robust multi-agent Bellman operator, RoMEX-$φ$ enables tractable worst-case value estimation from nominal interaction data using a centered empirical robust discrepancy. We introduce the robust Multi-Agent Decoupling Coefficient (robust MADC) to characterize the intrinsic exploration complexity arising from strategic interactions and adversarial transition uncertainty. We establish sublinear robust regret guarantees governed by the robust MADC rather than explicitly by the state and joint action space sizes, replacing tabular dependence with intrinsic function-class complexity. Numerical experiments on a scalable general-sum DRMG under total variation uncertainty show that RoMEX-$φ$ is substantially more resilient to transition shifts than its non-robust counterpart while remaining competitive with an exact tabular robust baseline. Our results provide a scalable framework for distributionally robust multi-agent reinforcement learning with general function approximation.

Abstract: In the A* search algorithm, the tie-breaking strategies for nodes with the same $f$-value determines which states A* expands on the final $f$-layer. For nine standard tie-breaking strategies, we show that under a consistent heuristic, every pair has positive-cost instances favoring each strategy over the other by an arbitrarily large additive expansion gap. A parameterized unit-cost grid example also gives unbounded expansion-count ratios between low-$h$ with FIFO and LIFO. In unit-cost search with $h > 0$ at non-goals, exact heuristic values near the goal lead to complementary extremal results: low-$h$ minimizes the number of remaining expansions from a common configuration within the perfect region, while high-$h$ maximizes the total number of expansions when every final-plateau state with $h=1$ is a goal predecessor. Finally, with the evaluation function $f_α = g + αh$, when $h>0$ at non-goals, every heuristic weight $0 \leq α<1$ eliminates tie-breaking sensitivity, and all tie-breaking strategies expand the same set of states.

Abstract: Multi-turn agents trained with reinforcement learning (RL) receive a single scalar reward per trajectory, which motivates self on-policy distillation (OPD) to supply dense token-level supervision from a self-teacher with privileged task skills, letting a skill-free student internalize them. This recipe, however, is undermined by two findings in agentic tasks: privileged information alone does not always make a teacher reliable, and the benefit of teacher supervision is stage-dependent. We therefore propose RetireOPD (Self-Retiring On-Policy Distillation), which first optimizes a decoupled, skill-conditioned teacher with environment rewards and then trains a skill-free student jointly with RL and OPD. Rather than following a predefined distillation schedule, RetireOPD adopts Adaptive Retirement: the student drops the teacher on its own once their discrepancy stops shrinking and it reaches a target fraction of the teacher's success rate, after which training proceeds with RL alone. Across Qwen2.5 models from 1.5B to 7B, RetireOPD improves ALFWorld success rate over RL baseline by 14.1% to 18.8% and WebShop accuracy by 11.8% to 19.0%, and surpasses its own skill-conditioned teacher in every setting.

Abstract: Ad Hoc Teamwork (AHT) addresses the challenge of designing agents capable of coordinating with novel partners without prior coordination. However, progress in the field is hindered by the prohibitive computational cost of the AHT research lifecycle, the lack of standardized benchmark implementations, and the absence of a diverse, validated evaluation teammate suite. In this work, we introduce JaxAHT, the first open-source, JAX-based library designed to accelerate and standardize the AHT research lifecycle. Leveraging JAX's hardware acceleration and massive parallelization capabilities, JaxAHT provides a unified framework for teammate generation, ego agent training, and evaluation against unseen teammates, achieving approximately 95x wall-clock speedup over PyTorch counterparts. Alongside the library, we contribute a diverse suite of evaluation teammates across the domains of Level-Based Foraging, Overcooked, and Hanabi. To illustrate the value of the framework, we use it to conduct a large-scale, compute-controlled benchmark study comparing teammate generation and AHT agent learning methods, finding that no algorithm consistently performs best, and that agent modeling primarily offers benefits in role-based scenarios with diverse teammates.