NeurIPS 2026 论文列表

Recent multimodal large language models (MLLMs) can perceive a traffic scene, reason about it, and propose a driving action within a single model. However, existing MLLM-based drivers reason from scratch at each frame and commit to a single explanation, discarding earlier hypotheses about the scene. When the scene is ambiguous, the agent locks in on the wrong interpretation and acts on it before evidence accumulates. We propose DriveMind, an MLLM-based driving framework that keeps several hypotheses about the scene alive at the same time and updates their weights as new frames arrive. Inspired by particle filtering, DriveMind propagates each hypothesis through a VLM, scores it against the current observation, and resamples to maintain diversity. On Bench2Drive, DriveMind raises the Driving Score (DS) from 71.36 to 81.37 and the Success Rate (SR) from 45.45% to 54.09%, with the largest gains on ambiguous scenarios such as unsignalized intersections (+20.10 DS) and oncoming-traffic interactions (+20.50 DS). Beyond the overall metrics, DriveMind also improves fine-grained interactive driving abilities such as merging, emergency braking, and give-way. The code will be released soon.

Existing latent world models for autonomous driving have opened a promising path toward future-aware driving intelligence. However, they typically treat future latent states as prediction targets or auxiliary signals, rather than directly conditioning trajectory planning. This can entangle current and future features in latent space. In this work, we propose DriveFuture, a future-aware latent world modeling framework for autonomous driving that explicitly learns planning-oriented foresight by conditioning the current latent state modeling process on future world states. Specifically, during training, the model first predicts future latent world states from the current latent state and ego action, and then refines the prediction against the ground-truth future latent state via cross-attention. The resulting future-aware latent serves as an explicit condition for a diffusion-based trajectory planner. During inference, DriveFuture conditions on the predicted future latent state instead of the ground-truth future state. DriveFuture achieves SOTA performance on the public NAVSIM benchmarks, reaching 55.5 EPDMS on NAVSIM-v2 navhard, 89.9 EPDMS on NAVSIM-v2 navtest, and 90.7 PDMS on NAVSIM-v1 navtest, respectively. These results suggest that the key to latent world modeling lies not merely in simulating future states, but more importantly in conditioning current decision-making on future states. Notably, as of April 2026, DriveFuture ranks 1st on the NAVSIM-v2 navhard leaderboard and achieves SOTA performance on NAVSIM-v1 navtest.

Recently, world-action models (WAM) have emerged to bridge vision-language-action (VLA) models and world models, unifying their reasoning and instruction-following capabilities and spatio-temporal world modeling. However, existing WAM approaches often focus on modeling 2D appearance or latent representations, with limited geometric grounding, an essential element for embodied systems operating in the physical world. We present DriveDreamer-Policy, a unified driving world-action model that integrates depth generation, future video generation, and motion planning within a single modular architecture. The model employs a large language model to process language instructions, multi-view images, and actions, followed by three lightweight generators that produce depth, future video, and actions. By learning a geometry-aware world representation and using it to guide both future prediction and planning within a unified framework, the proposed model produces more coherent imagined futures and more informed driving actions, while maintaining modularity and controllable cost. Experiments on the Navsim v1 and v2 benchmarks demonstrate that DriveDreamer-Policy achieves strong performance on both planning and world generation tasks. In particular, our model reaches 89.2 PDMS on Navsim v1 and 88.7 EPDMS on Navsim v2, outperforming existing world-model-based approaches while producing higher-quality future video and depth predictions. Ablation studies further show that depth learning provides complementary benefits to video imagination and improves planning robustness. Our code is available to facilitate further research.

Joint-Embedding Predictive Architectures (JEPAs) are prone to representation collapse, typically mitigated through empirical heuristics. We develop an early-training stability theory that unifies these heuristics. Linearising the coupled JEPA gradient flow around the trivial fixed point reveals two competing effects: a driving force ($\gamma$) and a decay effect ($\sigma$). Under approximate spectral decoupling, a per-mode stability ratio $\mu_i = \gamma_i / \sigma_i$ factorises into independent data-side and predictor-side terms and the count of unstable modes tracks the rank of representations that can emerge. The framework predicts a phase boundary, which we confirm empirically across more than 800 Tabular-JEPA configurations. It also unifies predictor scaling, masking ratio, and EMA as distinct mechanisms for shifting $\mu$. Guided by this analysis, we introduce ResidualPred, a transformer predictor whose attention is biased toward the identity at initialisation; it improves both effective rank and downstream accuracy on tabular benchmarks and on I-JEPA pretraining over CIFAR-10, CIFAR-100, and STL-10. Our framework connects empirical collapse-avoidance heuristics to an explicit dynamical picture, yielding theory-driven stabilizers. Code is available as supplementary material.

Drifting models have recently gained attention for generating high-quality samples in a single forward pass. During training, they learn a push-forward map by following a vector-valued field, the *drift field*. We ask whether this procedure is equivalent to optimizing a scalar loss and find that, in general, it is not: drift fields are *not conservative* and cannot be written as the gradient of any scalar potential. We identify the position-dependent normalization as the source of non-conservatism, with the Gaussian kernel as the unique radial exception. Guided by this, we introduce the *sharp kernel* $k^\\#$ and a sharp-normalized drift field that is conservative for general radial kernels. The resulting vector field is the gradient of a scalar potential that can be optimized directly using stochastic gradient descent. Moreover, the field has the form of a score difference of kernel density estimates, and gives exact equilibrium identifiability. Thus, sharp normalization closes the gap to related literature, such as Wasserstein gradient-flows and denoising score matching, also for non-Gaussian kernels. Empirically, sharp normalization preserves the performance of the original drifting objective, suggesting that the non-conservative flexibility is not required for high-quality generation.

We propose Drifting Field Policy (DFP), a non-ODE one-step generative policy built on the drifting model paradigm. Under online RL finetuning, we frame the policy update as a reverse-KL Wasserstein-2 gradient flow toward a soft target policy, realized by DFP as a steepest-descent step on probability space. By construction, this gradient step is decomposed into an ascent toward higher action-value regions and a score matching with the anchor policy as a trust region. We further derive a simple, tractable surrogate of the otherwise intractable update loss, akin to behavior cloning on top-K critic-selected actions. We find empirically that this mechanism uniquely benefits the drifting backbone owing to its non-ODE parameterization. With one-step inference, DFP achieves state-of-the-art performance on several manipulation tasks across Robomimic and OGBench, outperforming ODE-based policies.

Generating neural network weights in a single forward pass promises to amortize model training for fast sampling, transfer, and model-set construction. Yet accuracy alone is insufficient: existing generators can map diverse latent inputs to weights that remain functionally close to their training checkpoints. We study this functional mode collapse in the one-step regime and introduce DriftWeight, a MeanFlow-based generator with temperature-scaled repulsive drift in a checkpoint-augmented PCA proxy space. The method uses structured negatives from the current batch and optimization history to push generated weights away from observed clusters while preserving task performance. Our key finding is that repulsion is not automatically compatible with one-step inference. In a controlled ablation, applying the drift loss at the same boundary evaluation used for sampling collapses one-step accuracy even though multi-step integration remains accurate. DriftWeight mitigates this boundary-localized failure with a consistency objective that anchors the inference point to the training-weight manifold. Across MNIST, Fashion-MNIST, and CIFAR-10, DriftWeight matches or exceeds reported 100-step DeepWeightFlow accuracy in a single forward pass. On MNIST, where we directly evaluate error-set functional overlap, it reduces MaxIoU from the DeepWeightFlow regime of approximately $0.82$ to $0.66$. On CIFAR-100 with a ViT-Base target ($\sim$86M parameters), one-step samples remain within $1.7$ percentage points of the training-population mean.

We propose Drift-Resistant Navigation World Model, a generative model that mitigates both perceptual drift and geometric drift in conventional rollout-based navigation world models. Existing methods recursively feed generated content into subsequent steps, causing noise accumulation and degraded predictions, i.e., perceptual drift. Meanwhile, their predictions often deviate from the agent’s motion, resulting in geometry drift. We address both types of drift by redesigning world-model prediction as an anchor-guided rollout. Instead of rolling out every frame sequentially, we first predict sparse future anchors that serve as stable long-range targets, and then generate intermediate frames within each chunk conditioned on both past context and future anchors. Importantly, these sparse anchors also provide geometric constraints, supported by bidirectional epipolar geometry, to localize where corresponding content should appear in the intermediate frames. Experiments on four benchmarks demonstrate consistent improvements over strong baselines in long-horizon visual quality, geometric consistency, and multi-view coherence. These gains further translate into improved downstream planning performance under the same planners, highlighting the importance of drift-resistant, geometry-aware prediction for reliable navigation world models.

Mapping reaction pathways and transition states (TS) is fundamental to chemistry but computationally expensive at scale. The minimum energy pathway (MEP) dictates reaction rates and mechanisms, yet recovering it via electronic-structure methods requires thousands of costly force evaluations. Recent generative models accelerate TS identification but require slow iterative inference and only predict isolated saddle-point snapshots, missing the continuous reaction trajectory. We introduce Drift-React, an $\mathrm{SE}(3)$-equivariant generative framework that predicts complete reaction pathways in a single forward pass from only reactant and product geometries. By shifting distribution evolution to training via a Sinkhorn-weighted drifting field, Drift-React eliminates both the iterative force evaluations of NEB-style methods and the sequential ODE/SDE integration of diffusion and flow matching models. Evaluated on the Transition1x and Halo8 datasets, our one-step model generates physically consistent MEPs that accurately capture energetic bottlenecks and enable arbitrary-resolution sampling along the reaction coordinate. For isolated TS prediction, Drift-React matches the sub-Ångström accuracy of state-of-the-art iterative models while delivering orders-of-magnitude acceleration, clearing a major computational bottleneck for large-scale reaction network exploration.

Understanding user preferences from noisy and temporally evolving social media behaviors is fundamentally challenging due to interest drift, where user preferences shift across time and exhibit both multi-scale temporal patterns and diverse co-existing interests. To address this, we propose DUMoE, a unified framework for drift-aware multimodal user representation learning. Our model consists of (i) a temporal dynamics-aware backbone that captures and integrates static profiles, short-term behavioral signals, and long-term dependencies into a coherent representation, and (ii) a sparse mixture-of-experts (MoE) interest adapter that disentangles multiple latent interests via expert specialization and adaptive routing. Each expert models a distinct interest subspace, while a gating network dynamically selects and aggregates a sparse subset of relevant experts for each user. To enable stable and effective optimization, we further introduce a three-stage training strategy that decouples backbone learning, expert specialization, and gating optimization. Extensive experiments on real-world social media datasets show that DUMoE consistently outperforms state-of-the-art methods on both user interest prediction and interaction prediction tasks.

Offline reinforcement learning requires improving a policy from fixed data while avoiding out-of-distribution actions with unreliable value estimates. Diffusion and flow policies handle this trade-off by modeling the behavior distribution to regularize the RL objective, but they require iterative denoising, solver integrations, and in more efficient variants, distillation or other approximations at inference. We propose DriftQL, which combines a drift-based behavioral regularizer with critic-driven policy improvement. The value signal biases the policy toward high-value regions of the data support, while attraction and repulsion together keep generated actions near the data and prevent collapse onto a single mode. DriftQL is implemented as a single network with a unified training objective and generates actions in a single forward pass. On D4RL and OGBench, DriftQL consistently outperforms diffusion and flow methods, advancing the state of the art. Under degraded data quality, where the baselines visibly struggle, DriftQL remains close to its clean-data performance, positioning it as a promising alternative to diffusion and flow-based methods while maintaining the simplicity and efficiency of deterministic approaches.

Modern image editors excel at semantic manipulation and visual synthesis, yet remain limited in precise spatial control, motivating the development of drag-based editing. However, existing drag-based methods often struggle to balance drag accuracy with natural, plausible, and intent-aligned generation. We propose MoRe-Drag, a motion-grounded drag-based editing method. Our key insight is to treat pixel-space warping as coarse motion evidence, and to inject this evidence into the generative sampling trajectory. Specifically, MoRe-Drag performs region-aware latent recomposition over refinement, inpainting, and anchor regions, coupled with stage-adaptive conditioning that progressively shifts from motion-grounded structure formation to semantic refinement. We further support an instruction-free interface by adapting the MLLM-based text encoder for drag-aware instruction inference. Experiments on DragBench-SR and DragBench-DR show that MoRe-Drag substantially improves drag precision over strong base editors and achieves superior drag accuracy among SOTA drag-based methods, while delivering strong semantic consistency and visually realistic results. Code and dataset will be publicly released.

While one-step text-to-image generators, a fast and deployment-friendly class of visual generative models, can produce samples with a single forward pass, existing preference-finetuning methods fail to simultaneously achieve efficient adaptation, reward-model agnosticism, and diverse generation. In this work, we propose Drifting Preference Optimization (DrPO), an online preference-finetuning method for one-step generators. Inspired by the recent work of Drifting Models, DrPO first uses the target reward model to construct an on-policy dipole reward model from positive and negative samples. This dipole model induces a preference drifting field in latent feature space, whose gradients are then used to optimize the generator without backpropagating through the target reward model. Empirically, we evaluate DrPO across multiple reward models and SD-Turbo and SDXL-Turbo one-step generators, including HPSv3 and GenEval as representative alignment benchmarks, and validate that DrPO can efficiently and effectively finetune one-step generative models. Preliminary experiments further suggest that this sample-based gradient synthesis can be extended to offline preference finetuning.

Multi-agent LLM systems improve reasoning and tool use via role specialization, yet reinforcement learning (RL) post-training for such systems remains unstable and underexplored. We theoretically pinpoint a key source of instability when extending group-based RL to cooperative multi-agent LLM systems: under GRPO-style optimization, a global normalization baseline can mismatch heterogeneous agents' reward distributions, inducing gradient-norm instability. Based on this finding, we propose Dr. MAS, a simple and stable RL recipe for cooperative multi-agent LLM systems. Algorithmically, Dr. MAS normalizes advantages per agent using each agent's own reward statistics, which calibrates gradient scales and dramatically stabilizes the training. Systemically, Dr. MAS provides an end-to-end multi-agent RL framework with scalable orchestration, flexible per-agent LLM serving and optimization, and shared resource scheduling of actor backends. Unlike single-actor frameworks such as veRL, Dr. MAS natively supports multiple heterogeneous LLMs over a unified GPU pool. Lifecycle-aware backend management and dynamic dispatch release inactive models' GPU memory and schedule active agents on demand, enabling hardware-efficient co-training. We evaluate Dr. MAS on multi-agent math reasoning and multi-turn search with Qwen2.5 and Qwen3. Dr. MAS achieves clear gains over vanilla GRPO (e.g., +5.6\% avg@16 and +4.6\% pass@16 on math, and +15.2\% avg@16 and +13.1\% pass@16 on search) while largely eliminating gradient spikes. It also remains highly effective under heterogeneous agent-model assignments while improving efficiency. Code is available at https://anonymous.4open.science/r/DrMAS-0787.

Understanding human kinematics in outdoor, real-world environments is critical for assistive robots and intelligent infrastructure. Many perception systems rely on cameras and LiDAR, which infer motion indirectly from spatial observations, making estimates sensitive to discontinuities and ambiguity under low lighting or at distance. In contrast, mmWave radar is robust to lighting and weather, operates at long range, and directly measures motion through Doppler signatures. However, Doppler-based kinematic understanding remains largely unexplored in the wild because existing datasets are primarily controlled and scripted. We introduce DopplerWild, a dataset and benchmark for evaluating Doppler-based kinematic understanding in the wild. DopplerWild comprises 447k radar frames across four outdoor locations, including an unlabeled subset for self-supervised representation learning and a labeled subset of 539 subjects with nearby-person interference (56\%) and occlusion (16\%). The benchmark spans coarse and fine-grained motion classification, as well as velocity estimation. Beyond aggregate metrics on controlled datasets, DopplerWild evaluation isolates real-world factors: subject variation, location shift, multi-person interference, occlusion, sensing geometry, and low-label regimes. DopplerWild reveals three findings. First, self-supervised pretraining matches supervised performance with less than half the labeled data and improves performance under interference, occlusion, and lateral motion. Second, we identify key challenges for single-view Doppler sensing, including physical limits in lateral motion, subtle asymmetric limb motion, and location distribution shifts. Third, we observe bidirectional transfer asymmetry between DopplerWild and an external controlled Doppler dataset, suggesting that real-world coverage is important for transferable Doppler representations. DopplerWild provides an evaluation framework for characterizing the generalization, limits, and transferability of Doppler-based kinematic representations in the wild.

Top-1 accuracy on ImageNet-1k remains the most universally reported metric in visual recognition, despite its well-documented label quality issues. In the paper, we describe the effort to obtain accurate and complete ImageNet-1k validation set annotations. The result -- ReImageNet -- is a from-scratch reannotation, which includes multilabel correction, object localisation, revised class definitions, and semantic attributes---text-recognition, rendition, reflection, crowd, and dominant. The reannotation reveals that $\approx13$\% of the original ImageNet-1k labels are incorrect, $32.7$\% of images are multilabel and $4.7$\% contain no object from an ImageNet-1k class. With the new labels, top-1 accuracy increases by $0.7$--$1.4$\% for supervised models and by $6$--$8$\% for MLLMs. We show that annotation errors in ImageNet-1k propagate into its derivative test sets, indicating that the problem is structural rather than specific to any single benchmark. We observed that LLMs outperform unaided crowd annotators, and that human and LLM collaboration with appropriate tooling represents the current quality ceiling for annotation at this scale. All [annotations](https://huggingface.co/datasets/c1rcuslegend/ReImageNet), [class definitions](https://gonikisgo.github.io/imagenet-annotation-preview/), [guidelines](https://gonikisgo.github.io/imagenet-annotation-preview/), and [analysis code](https://github.com/klarajanouskova/ImageNet) have been publicly released.

Reinforcement-learning-based post-training for large language models vastly improves capabilities on verifiable tasks however it suffers from training instability and diversity collapse. As a solution, many new advantage functions have emerged. Each method simultaneously changes which problems receive gradient, the balance between positive and negative updates, and the overall gradient scale making their comparison difficult. We propose a unifying framework that decomposes each function by their induced positive and negative gradient mass. Analyzing policy performance and the geometry of model weights reveals: symmetric methods are better but slow and asymmetric methods are fast. Combining the advantages of both, we build the FADE: Focal Adaptive Dynamic Entropy advantage which is fast, diverse and accurate at various model scales (7B, 32B).

Large language models spend enormous parameter capacity approximating arithmetic, a deterministic function with an exact solution, and still fail reliably on multi-digit operations. We propose Arithmetic Residual Blocks (ARBs): frozen, differentiable modules inserted into the transformer forward pass that compute exact integer arithmetic using Residue Number System (RNS) encoding on unit circles. The model learns only the interface, a gated injection pathway and a low-rank adapter on the language model head, while the base model and all arithmetic computation remain frozen. The architecture embodies a design principle we call Don't Learn What You Can Compute: any deterministic function expressible as tensor operations can be embedded as a frozen residual block, and the model will learn to route through it because the gradient reward for exact answers dominates internal approximation. We validate this principle by inserting ARBs into a frozen SmolLM2-360M model, training only 1.7M parameters (0.47% of the base model). On exact-match evaluation across addition, subtraction, multiplication, and division with operands up to 3 digits, the augmented model achieves 99.9%+ accuracy across all operations, compared to 4.6% mean accuracy for the unmodified base model. A logit analysis of the frozen base model confirms that the injection pathway overrides the base model's prior uniformly across operations, and that the residual errors (0.047%) concentrate at later digit positions in autoregressive generation, not at any particular operation. These findings motivate the hypothesis that pre-training with ARBs, where no misaligned frozen representations exist, would yield ceiling performance across all operations.

Standard embodied evaluations do not independently score whether an agent correctly commits to task completion at episode closure—a capacity we call terminal commitment. Behaviorally distinct failures—never completing the task, completing it but failing to stop, and reporting success without sufficient evidence—collapse into the same benchmark failure. We introduce VIGIL, an evaluation framework that makes terminal commitment independently measurable. Under VIGIL’s default protocol, agents observe only egocentric RGB, receive no action-success signals, and must end each episode with a semantic report checked deterministically against hidden world state. This yields two separate scores: world-state completion (W) and benchmark success (B), where B additionally requires a correct terminal report. This decoupling makes four outcome categories distinguishable: missed execution, post-attainment drift, unsupported commitment, and verified success. Across 20 models on 1,000 frozen episodes, systems with comparable W differ by up to 19.7 pp in B: one model converts achieved states into correct reports, while another with near-identical execution drifts past the goal without closing. An action-feedback intervention further tests the separation: execution-oriented signals improve W broadly, yet commitment failures persist in models that do not already ground terminal reports in the achieved state. VIGIL provides a protocol that makes terminal commitment independently visible and scorable.