arXiv subjects

Haofei Sun

Publications and source records attributed to Haofei Sun.

8 recordsLinked to original sources

arXiv2026

Geometry Conditioning in an Embodied SLM: Training Controls and Robustness Diagnostics in a 0.8B Hybrid Model

We study how physical-state inputs affect a 0.8B hybrid language model adapted for manipulation with 6.2M trainable parameters. Six conditions are trained on three LIBERO-Spatial tasks and evaluated over three seeds and 540 held-out rollouts. Conditioning recurrent decay gates on geometric increments yields 28.9% success, compared with 36.7% when those increments are shuffled during training and 24.4% without explicit object/goal geometry. Both geometry policies receive correct inputs at evaluation. A token adapter using the same increments scores 27.8%; differences vary across seeds and remain inconclusive. Token-clock conditioning scores 11.1%, including one seed that fails to converge. In separate robustness tests, a state-only relative-coordinate policy retains 7/10 success under frame relabeling, whereas all four tested visual policies fall to at most 3/20 after a 5 cm object displacement. These results show no reliable advantage from training-time geometric alignment under this recipe and illustrate the gap between coordinate invariance and physical-layout generalization. Episode records, seed-level analyses, and figure-generation code accompany the paper.

cs.RO↗

arXiv2026

When Memory Updates but Behavior Does Not: Repairing Implicit Stale Dependencies in Personalized Agent Responses

Memory-augmented agents can know that a user's stored state is outdated and still plan around the old value. The STALE benchmark calls this the implicit policy adaptation (IPA) gap. We identify one structural contributor: draft-anchored verification checks what a response says, and in an open-ended response the stale dependency is usually unsaid. StateAuditor therefore audits in the opposite direction, from stored state to draft. An LLM proposes candidate old-to-new transitions from timestamped evidence; deterministic code pins each quotation to a single entry, checks that the new evidence really is newer, and lets only these verified transitions trigger repair. What is verified is provenance and chronology - not semantic supersession. On STALE's full protocol (400 scenarios, 50-session histories, one independent response per query), strict single-query VTA scores .736 against .686 for our locked predecessor under the same judge: a +5.0-point paired gain (95% CI [+2.9, +7.2]) coming almost entirely from IPA and premise resistance (PR). The benchmark's own judge, from a third model family, reproduces the gain (.738 vs. .680). On an independent cross-family preference-evolution benchmark (HorizonBench), the full draft-audit-repair pipeline over a gold-derived structured store raises current-preference accuracy (user-clustered p<.01), though a matched control shows most of this external gain is the draft-side audit itself; a harder authored lifecycle set gives no gain, bounding the claim while false invalidation stays controlled. On STALE, by contrast, a matched control (same evidence, adapter, and call budget) scores only .692 (+0.6 over the predecessor, n.s.), attributing the STALE gain to the transition machinery rather than added context or calls. We make no claim about general-purpose agent memory.

cs.AI↗

arXiv2026

A plug-and-play generative framework for multi-satellite precipitation estimation

Reliable precipitation monitoring is essential for disaster risk reduction, water resources management, and agricultural decision-making. Multi-source satellite observations, particularly the combination of geostationary infrared and passive microwave measurements, have become a primary means of precipitation detection. Traditional multi-source satellite precipitation estimation methods remain computationally inefficient, and many deep learning methods lack the flexibility to incorporate new sensors without retraining the full model. Here we introduce PRISMA (Precipitation Inference from Satellite Modalities via generAtive modeling), a plug-and-play latent generative framework for multi-sensor precipitation estimation. PRISMA learns an unconditional precipitation prior from IMERG Final fields and constrains it through independently trained, sensor-specific conditional branches, allowing new observation sources to be incorporated without retraining the generative backbone. Applied to FY-4B AGRI infrared and GPM GMI microwave observations, PRISMA improves Critical Success Index by up to 40.3% and reduces root-mean-square error by 22.6% relative to infrared-only estimation within microwave swaths, while also improving probabilistic skill and maintaining an average inference time of about 37 s. Independent rain-gauge validation across China confirms consistent gains, and typhoon case studies show that microwave conditioning restores eyewall and spiral rainband structures, reducing storm-core mean absolute error by up to 42.3%. PRISMA thus provides an extensible and efficient framework for multi-sensor precipitation estimation.

physics.ao-ph↗

arXiv2026

A Data-Driven Regional Model for Skillful Medium-Range Typhoon Prediction

Accurate prediction of tropical cyclones remains a major challenge for both numerical weather prediction and emerging artificial intelligence weather prediction systems. While recent global AI models have demonstrated strong skill in large-scale circulation prediction, they often struggle to represent the mesoscale structures critical for tropical cyclone intensity and precipitation. Here we develop the Hybrid Intelligent Typhoon System (HITS), a regional AI forecasting framework for medium-range typhoon prediction over the Asia-Pacific region, trained on a newly constructed 9-km high-resolution typhoon reanalysis dataset. The model combines regional autoregressive prediction with large scale dynamical constraints from the state-of-the-art ECMWF Artificial Intelligence Forecasting System (AIFS), allowing it to remain dynamically consistent with the evolving large-scale circulation while resolving mesoscale structures. HITS is further extended with a structure-aware perceptual training strategy (HITS-LPIPS) that improves the representation of convective and typhoon rainband structures. Experiments show that the hybrid framework substantially improves precipitation structure and typhoon intensity forecasts compared with both purely autoregressive regional AI models and standalone AI downscaling approaches. In particular, HITS-LPIPS reduces intensity errors by up to 47.8% relative to AIFS at a 72 hour lead time and produces a near-unbiased wind-pressure relationship for simulated typhoons. These results demonstrate that dynamically constrained regional AI systems provide a promising pathway for improving medium-range typhoon prediction.

physics.ao-ph↗

arXiv2026

ProactiveMobile: A Comprehensive Benchmark for Boosting Proactive Intelligence on Mobile Devices

Multimodal large language models (MLLMs) have made significant progress in mobile agent development, yet their capabilities are predominantly confined to a reactive paradigm, where they merely execute explicit user commands. The emerging paradigm of proactive intelligence, where agents autonomously anticipate needs and initiate actions, represents the next frontier for mobile agents. However, its development is critically bottlenecked by the lack of benchmarks that can address real-world complexity and enable objective, executable evaluation. To overcome these challenges, we introduce ProactiveMobile, a comprehensive benchmark designed to systematically advance research in this domain. ProactiveMobile formalizes the proactive task as inferring latent user intent across four dimensions of on-device contextual signals and generating an executable function sequence from a comprehensive function pool of 63 APIs. The benchmark features over 3,660 instances of 14 scenarios that embrace real-world complexity through multi-answer annotations. To ensure quality, a team of 30 experts conducts a final audit of the benchmark, verifying factual accuracy, logical consistency, and action feasibility, and correcting any non-compliant entries. Extensive experiments demonstrate that our fine-tuned Qwen2.5-VL-7B-Instruct achieves a success rate of 19.15%, outperforming o1 (15.71%) and GPT-5 (7.39%). This result indicates that proactivity is a critical competency widely lacking in current MLLMs, yet it is learnable, emphasizing the importance of the proposed benchmark for proactivity evaluation.

cs.AI↗

arXiv2026

StormDiT: A generative AI model bridges the 2-6 hour 'gray zone' in precipitation nowcasting

Accurate short-term warnings for extreme precipitation are critical for global disaster mitigation but are hindered by a persistent predictability barrier at the 2-6 hour horizon -- the "nowcasting gray zone." In this window, traditional observation-based extrapolation fails due to error accumulation, while numerical weather prediction is computationally too slow to resolve storm-scale dynamics. Recent generative AI approaches attempt to bridge this gap by decomposing precipitation into separate deterministic advection and stochastic diffusion components. However, this decomposition can sever fundamental causal links between entangled atmospheric processes, such as the dynamic initiation of convection triggered by boundary advection. Here we present StormDiT, a unified generative model that treats weather evolution as a holistic spatiotemporal problem, learning the coupled physics of the gray zone without human-imposed structural priors. Trained on a massive dataset of 7,720 precipitation events from China, our model achieves a breakthrough in long-horizon stability. On a heavy-rainfall test set, it maintains skillful prediction for strong convection ($\ge$ 35 dBZ) with a Critical Success Index (CSI) near 0.2 across the full 6-hour forecast at 6-minute resolution. Crucially, the model exhibits superior probabilistic calibration, accurately quantifying operational risks. On the public SEVIR benchmark, our unified paradigm more than doubles the state-of-the-art 1-hour performance for heavy rain and establishes the first robust baseline for 3-hour forecasting. Furthermore, interpretability analysis reveals that the model attends to non-local physical precursors, such as outflow boundaries, explicitly validating its emergent understanding of convective organization.

physics.ao-ph↗

arXiv2025

Nowcast3D: Reliable precipitation nowcasting via gray-box learning

Reliable nowcasting of extreme precipitation remains difficult because convective systems are strongly nonlinear, multiscale, and nonstationary in 3D. Radar is the backbone of nowcasting, yet existing methods struggle to predict extremes: physics-based extrapolation cannot capture growth and decay, deterministic learning tends to oversmooth and underestimate peaks, and purely generative models often lack physical consistency. Hybrid schemes help but are mostly limited to 2D composite reflectivity, collapsing the atmosphere into one layer and discarding vertical structure critical for height-dependent dynamics. We introduce Nowcast3D, a gray-box, fully 3D framework that works directly on volumetric radar reflectivity. The end-to-end model couples physically constrained neural operators (advection, local diffusion, and microphysics) with a conditional diffusion model to generate ensemble forecasts with quantified uncertainty. Trained on provincial-scale 3D volumes over a $10.24^\circ \times 10.24^\circ$ region and fine-tuned on a $2.56^\circ \times 2.56^\circ$ city region ($0.01^\circ \approx 1$ km), Nowcast3D provides near-real-time forecasts up to 3 h and outperforms competitive baselines in cross-region and temporal out-of-sample tests. It can also infer wind fields without labeled supervision, supporting physically plausible transport. In a nationwide blind evaluation by 160 meteorologists, Nowcast3D ranked first and was preferred in 57% of post-hoc assessments, surpassing the leading baseline (27%). These results highlight its reliability and operational value for extreme precipitation nowcasting.

cs.LG↗

arXiv2025

Intelligent Shanghai Typhoon Model (ISTM): A generative probabilistic emulator for typhoon hybrid modeling

To address the systematic underestimation of typhoon intensity in artificial intelligence weather prediction (AIWP) models, we propose the Intelligent Shanghai Typhoon Model (ISTM): a unified regional-to-typhoon generative probabilistic forecasting system based on a two-stage UNet-Diffusion framework. ISTM learns a downscaling mapping from 4 years of 25 km ERA5 reanalysis to a 9 km high resolution typhoon reanalysis dataset, enabling the generation of kilometer-scale near-surface variables and maximum radar reflectivity from coarse resolution fields. The evaluation results show that the two-stage UNet-Diffusion model significantly outperforms both ERA5 and the baseline UNet regression in capturing the structure and intensity of surface winds and precipitation. After fine-tuning, ISTM can effectively map AIFS forecasts, an advanced AIWP model, to high-resolution forecasts from AI-physics hybrid Shanghai Typhoon Model, substantially enhancing typhoon intensity predictions while preserving track accuracy. This positions ISTM as an efficient AI emulator of hybrid modeling system, achieving fast and physically consistent downscaling. The proposed framework establishes a unified pathway for the co-evolution of AIWP and physics-based numerical models, advancing next-generation typhoon forecasting capabilities.

physics.ao-ph↗