Daily Digest 2026-07-27

ArXiv ID: 2607.21635

Authors: Pin Qian, Su Wang, Yihang Chen, Qiaolin Yu, Xiaoyuan Wang, Zhitong Guo, Zhicheng Wang, Junxian You

Abstract:

Personal agents maintain memories, learned skills, tool configurations, and policy state that evolve with each user. Existing agent benchmarks often evaluate these capabilities in isolation: tool benchmarks test invocation under fixed APIs, memory benchmarks test recall or forgetting, and safety benchmarks test static policy compliance. We argue that personal-agent evaluation requires a different protocol: replaying the same temporal intervention across different persistent user-conditioned states and measuring how failures propagate across agent components. We formalize this requirement as four conditions: explicit temporal intervention, persistent state across the intervention, induced cross-dimensional effects, and variation in user-conditioned state. A focused audit of public benchmark protocols selected by explicit inclusion criteria identifies several close cases. Under our explicitly narrow operationalization, we did not find a protocol in that audited set satisfying all four conditions. This claim is scoped as a focused gap analysis with bounded literature coverage. This position paper proposes a minimal benchmark design and candidate reporting metrics for user-conditioned adaptation. The result is a concrete design requirement for future personal-agent evaluation, with metrics used as reporting tools for that requirement.

ArXiv ID: 2607.21653

Authors: Jian Hu, Huiying Li, Hao Zhang, Binfeng Xu, Yifan Zhang, Shaokun Zhang, Hemil Desai, Michael Demoret, Pavlo Molchanov, Jan Kautz, Yi Dong

Abstract:

Agentic reinforcement learning research is constant algorithm modification, new estimators, new pipeline stages, new rollout schemes, and in mainstream frameworks each change threads through layers of trainer, distributed backend, and rollout glue: the cost lands on the researcher at every iteration. Molt is a PyTorch-native training framework built to keep that cost small: a codebase compact and clean enough for a researcher to hold in their head, and for an AI coding assistant to read and reason about in its entirety, so the algorithm flow can be traced and changed end to end. The agent is an ordinary program, and one asynchronous loop trains multimodal and mixture-of-experts policies while never training on a token it did not generate, consistent in tokens, policy versions, and model semantics. Leanness does not cost performance: under a matched, fully asynchronous protocol, Molt is statistically comparable to a state-of-the-art Megatron-based stack. Molt is open source and provides recipes and containers at https://github.com/NVIDIA-NeMo/labs-molt.

ArXiv ID: 2607.21856

Authors: Ziran Yang, Chengshuai Shi, Raj Ghugare, Benjamin Eysenbach, Karthik Narasimhan, Chi Jin

Abstract:

Modern reasoning models depend on reasoning data, today sourced from human annotations or distilled from stronger LLMs. However, a rich and largely untapped source of supervision lies in expert systems (e.g., game engines, classical planners, theorem provers), which routinely produce near-optimal actions across diverse domains. But these experts are silent: they commit to an action without writing down the chain of thought (CoT) behind it. Recovering that CoT as natural-language reasoning would distill expert knowledge into a student that generalizes beyond the demonstrated actions. We treat it as a latent variable and study how to recover it from the action alone. Our approach, LeAct (Learning to reason from Actions), optimizes this latent variable: the student samples candidate CoTs for each expert action, and we retain those that measurably improve its own probability of recovering the action. Across imperfect-information games at multiple scales and a simulated robotics benchmark, LeAct reaches the solver's numerical floor on small enumerable games. At larger scale, it is $5\times$ closer to the solver than the strongest expert-iteration baseline. At Flop Hold'em ($\sim 10^9$ infosets), LeAct wins head-to-head by $+60$ mbb/g, and on the robotics probe it is the only training recipe that improves on direct imitation. We present a principled framework and the result: expert systems become a categorically new source of reasoning teachers for foundation models.

ArXiv ID: 2607.21946

Authors: Jiseok Kwak, Suhyeon Jo, Taewoo Kim, Yeongmin Kim, Byeonghu Na, Il-chul Moon

Abstract:

This technical report presents our approach to Challenge Track~3: SeePhys Pro at the 3rd AI for Math Workshop, where the task is to answer college-level physics questions whose statement and figure may be given partly or entirely as an image. Visual physics problems become substantially harder for large language models when the decisive information resides in a figure rather than in the text, and this modality gap widens as more of the problem migrates into the image. We address the task with a two-stage framework: a visual information extraction stage that re-expresses figure content as solver-readable text to close the modality gap, and a reasoning stage that orchestrates three heterogeneous solvers through multi-agent debate. Our analysis yields two findings: the gain from orchestration comes from reliable answer selection rather than from additional debate, and the value of a figure aid scales with how much of the problem is locked inside the image. The resulting pipeline improves overall accuracy over a single-agent baseline from 0.643 to 0.802 on the public split, and won 1st place on both the public and the private leaderboard (private overall 0.743).

ArXiv ID: 2607.21971

Authors: Shujin Wu, Cheng Qian, Xiusi Chen, Heng Ji

Abstract:

Test-time scaling through iterative self-evolution with environment feedback, as demonstrated by AlphaEvolve, shows remarkable performance gains. We hypothesize that the success of such evolution frameworks hinges on meta-skills, such as self-reflection with environment feedback, that enable effective multi-round refinement, yet are largely neglected by traditional post-training. To bridge this gap, we present MetaEvolve, a framework designed to develop these meta-skills via a data synthesis pipeline, evolution-aware reinforcement learning (RL), and inference-time evolutionary search. Concretely, we ground MetaEvolve in coding, where program execution provides natural, continuous reward signals beyond binary correctness. Building on these signals, we synthesize evolution trajectories as training data, each containing a current program, its fitness score (combining correctness and efficiency), and a history of prior attempts, and train the model via RL with verifiable rewards derived from test case execution. By training on large-scale code data, we aim to inspire generalizable domain-agnostic meta-skills that can transfer broadly to open-ended problems where such rich training signals are scarce. Across seven coding benchmarks, MetaEvolve outperforms the strongest baseline by 10.01% absolute on in-distribution tasks and 24.12% on out-of-distribution tasks. On open-ended algorithm optimization problems entirely outside the training domain, it further achieves a 46.9% relative improvement. These results demonstrate that explicitly cultivating self-evolution meta-skills offers a principled path toward more capable and autonomously self-evolving AI.

ArXiv ID: 2607.22430

Authors: Xiangteng Zhang, Yang Guan, Bo Zhang, Ya-Qin Zhang, Shengbo Eben Li

Abstract:

Learning world models that infer environment dynamics from high-dimensional observations and predict outcomes under candidate actions is central to planning and control. Joint-Embedding Predictive Architectures (JEPAs) provide a compelling framework for learning such models in representation space. Recent action-conditioned extensions perform promisingly in visual control and latent-space planning, but leave a fundamental question unresolved: when does controlled latent prediction identify both the underlying state and the controlled dynamics? This is challenging under nonlinear observations and behavior policies with limited conditional action variation, where state-dependent evolution and action effects can be statistically confounded. We establish a joint identifiability theory for controlled world models with Gaussian latent states under state-dependent Gaussian behavior policies. We identify two policy-dependent conditions: spectral separation of the predictable signal governs representation identifiability, while non-degenerate conditional action variation governs transition identifiability. We prove that when both conditions hold, every global minimizer of the JEPA objective identifies the latent state and controlled transition up to an orthogonal transformation. We further derive quantitative bounds on representation and transition identifiability under approximate optimization. Finally, we construct predictor perturbations along weakly excited action directions whose counterfactual-to-on-policy error ratio is the inverse transition-identifiability margin, revealing the cost of limited action coverage. Experiments across nonlinear observation maps and behavior policies corroborate the theory and demonstrate implications for transition identifiability, counterfactual prediction, and goal-conditioned latent planning.

ArXiv ID: 2607.21725

Authors: Liane Galanti, Dhruv Shah, Tri Dao

Abstract:

General-purpose robots need to reason about their actions, combining perception, world knowledge, planning, success detection, recovery, and low-level control. Today's state-of-the-art models attempt to combine all these capabilities into the learned policy via large-scale pre-training. Instead, we show that these capabilities can be decomposed into a general language-conditioned policy/control agent and a high-level agent manager/orchestrator. Rather than training policies to reason via pre-training, we build a closed-loop physical agent orchestrator that can do high-level planning, decompose the goal into achievable subgoals, command low-level motor commands, track and verify the outcome from low-level observations, and recover from failures. Our Physical Agency orchestrator (Pigey) can control existing vision-language-action (VLA) policies as well as parametrized skills to solve complex reasoning tasks in the real world, without any additional data collection or post-training. We evaluate Pigey extensively across simulation benchmarks and challenging real-world robotic manipulation tasks, and demonstrate significant performance improvements over existing generalist policies. On LIBERO-PRO, Pigey advances the state-of-the-art by over 4x (12.8% -> 53.3%) with no task-specific fine-tuning. On a real robot, Pigey lifts the frozen policy from near-zero to over 90% on reasoning-limited tasks. We call the difference between what frozen motor skills achieve alone and inside the agentic loop the orchestration gap.

ArXiv ID: 2607.22199

Authors: Lan V. Truong

Abstract:

Score-based diffusion models have achieved remarkable empirical success in generative modeling, yet their approximation-theoretic foundations remain incomplete. In particular, although classical universal approximation theorems guarantee that neural networks can approximate score functions, it remains unclear whether such approximation guarantees translate into approximation of the probability distributions generated by reverse diffusion processes. In this paper, we establish a rigorous quantitative connection between these two notions. Specifically, we prove that if a neural network approximates the true score function sufficiently accurately, then the probability distribution generated by the corresponding reverse diffusion model is close to the target data distribution in Kullback-Leibler (KL) divergence, up to an irreducible mismatch between the terminal distribution of the forward diffusion process and the prior used to initialize the reverse process. More precisely, we derive an explicit upper bound on the distribution approximation error in terms of the score approximation error, the diffusion noise schedule, and the terminal prior mismatch. Our analysis combines Hornik's universal approximation theorem, Girsanov's theorem on path space, and the data processing inequality for relative entropy. Complementary to recent work that studies score approximation under finite-sample statistical settings and structural assumptions on the data distribution, our work develops an approximation-theoretic analysis based on classical neural network approximation theory. The resulting theorem provides a simple and explicit guarantee linking neural network approximation of score functions to approximation of the probability distributions generated by reverse diffusion models.

ArXiv ID: 2607.21692

Authors: Jim Allchin

Abstract:

Sparse attention reduces the cost of long contexts by allowing each query to read only selected parts of the input. These selectors are often trained by distilling the attention patterns of a dense teacher, assuming that attention reveals which context the teacher actually uses. We test that assumption on retrieval tasks where the evidence for each answer is known exactly. By masking parts of the context and measuring whether the answer changes, we find that attention and causal dependence often disagree, and distilled selectors inherit the mismatch. Teachers attend to outdated facts they have learned to ignore, and their attention can vary across training runs even when they rely on the same evidence. In a two-step reference task, attention at the answer skips the intermediate step because it was resolved earlier in the forward pass: a selector trained on attention achieves 41% accuracy, while the same selector trained on causal evidence reaches 99% and matches the teacher. These evidence sets require no annotation: recovered from a frozen teacher by masking alone, they train selectors to the same accuracy. We find the same conflict in pretrained models: Qwen2.5-3B gives more attention to an outdated fact than the current one on 58% of conflicting-fact examples despite answering correctly, while Gemma-2-9B rises from 56% to 99% accuracy when restricted to the two relevant sentences. Attention shows where a model looks, not necessarily what its answer depends on; across the regimes we tested, that dependence matched or outperformed attention as a training target.

ArXiv ID: 2607.21752

Authors: Debarshi Kundu, Swaroop Ghosh, Vasant Honavar

Abstract:

Data-adaptive sparse attention masks substantially outperform fixed patterns (e.g., BigBird and Longformer) and can even exceed dense attention on long sequences. Existing adaptive approaches---including SBM-Transformer, Dynamic Mask Attention, and NSA---typically require additional learnable parameters, custom gradient estimators, or specialized CUDA kernels. We show that classical data compression provides an effective masking signal with \textbf{no additional parameters}. By computing per-block gzip compression ratios, we identify non-redundant content blocks and route long-range attention selectively through them. Intuitively, blocks that gzip cannot compress contain information not predictable from local repetition, making them natural long-range attention targets. Because the compression profile is input-dependent, the resulting sparse mask adapts dynamically to content without learned parameters, auxiliary losses, or custom kernels. On PG-19 byte-level language modeling at 92M parameters with 8K context, our method achieves 1.71 bits-per-byte (BPB), outperforming dense attention (2.89), BigBird (2.34), Longformer (3.21), and a reimplemented SBM-Transformer (3.38)---the only learned-mask baseline---by up to 1.67 BPB while adding no parameters. The advantage grows with sequence length, with the gap over BigBird widening from 0.05 BPB at 4K context to 0.63 BPB at 8K, while convergence is 3.3$\times$ faster.

ArXiv ID: 2607.21927

Authors: Anderson R. Santos

Abstract:

Full self-attention in large language models scales as O(N^2), which limits long-context document analysis to 65,536 tokens and requires costly GPU clusters. The Reduced Interaction Sampling (RIS) inference engine addresses this constraint as a model-agnostic architecture. Without modifying weights, RIS reduces self-attention complexity to O(N log N) using sparse stochastic geometry that fits within commodity memory limits. We validate RIS on Qwen2-1.5B-Instruct across two regimes. In controlled evaluations at 32,768 tokens (where native dense attention serves as the upper bound), RIS-Stochastic at 1% density and 70 ensemble seeds achieves 75.00% accuracy, outperforming the native dense baseline (71.88%), while RIS-Stochastic at 5% density and 10 seeds matches it (71.88%). This demonstrates that sparse attention acts as a regularizer: low density (1%) over multiple seeds filters out sequence-level noise, whereas higher density (5%) reintroduces distractor noise. Under the tightest budget, RIS-Structural reaches 68.75% accuracy at 1% density with just 10 seeds, recovering 75% of the contextual gap relative to the zero-context floor (59.38%). At 65,536 tokens, where dense attention triggers out-of-memory faults, RIS yields retrieval gains of up to 14.06 percentage points over the zero-context floor (51.56%), which is confirmed as marginally significant under McNemar's paired test (p = 0.078 < 0.10). All evaluations run on commodity, unaccelerated CPU servers (16-128 GB of RAM), demonstrating that long-context LLM inference is feasible on standard academic hardware without GPU acceleration.

ArXiv ID: 2607.22251

Authors: Wei Zhang, Xinwu Liu, Yihang Cheng

Abstract:

Low-Rank Adaptation (LoRA) is a widely used parameter-efficient fine-tuning method for large language models, but its performance depends strongly on how a fixed rank budget is distributed across Transformer modules. Existing adaptive-rank methods usually rely on local gradient statistics collected during training, which introduces extra memory and computation and overlooks task-conditioned global information flow. We propose IFCLoRA, a topology-aware rank allocation method applied before fine-tuning. Using a small calibration set and a frozen pretrained model, IFCLoRA builds a sparse task-conditioned interaction graph whose nodes represent LoRA-compatible modules. It combines a global information-flow topology prior with local gradient sensitivity to compute Information-Flow Centrality scores, which estimate each module's adaptation importance under multi-hop propagation. Ranks are then assigned once under a global budget. Across multiple models, tasks, and low-rank settings, IFCLoRA consistently outperforms LoRA, AdaLoRA, and EVA under matched training configurations and total rank budgets, while retaining training costs comparable to standard LoRA. On mathematical reasoning with LLaMA 3 8B, IFCLoRA improves over LoRA by 1.36 percent at rank 4 and 1.82 percent at rank 8. Further analysis shows task-dependent, non-uniform rank profiles, indicating that global information-flow structure provides an informative and interpretable prior for low-budget parameter-efficient fine-tuning.

ArXiv ID: 2607.22334

Authors: Hao Wang, Kun Yuan, Wenlin Zhong, Minglei Zhang, Han Xiao, Ming Sun, Honggang Qi

Abstract:

Open-weight language models from different families exhibit complementary capabilities, motivating their consolidation into a compact student through on-policy distillation (OPD). However, full-vocabulary OPD typically assumes a shared tokenizer, while existing cross-tokenizer methods may discard teacher probability mass or assign it to student tokens with unrelated content. We introduce Byte-Prefix Marginalization (BPM), which re-expresses the teacher's next-token distribution over the student vocabulary in a shared byte space. Specifically, BPM assigns each teacher token's probability to the longest student token whose byte representation is a prefix of the teacher token's bytes, aggregates mass mapped to the same student token, and places otherwise unmatched mass in an explicit residual category. This produces a vocabulary-complete, byte-aligned, and mass-preserving target for dense OPD. The target exactly recovers the teacher-induced byte-prefix marginal when the relevant prefix does not span multiple teacher tokens (a condition satisfied at more than 99% of training positions) and uses a mass-preserving, chain-factorized lower bound otherwise. Across Qwen3-32B, GLM-Z1-9B-0414, and MiniMax-M2.7 as teachers, BPM consistently outperforms current cross-tokenizer methods on six mathematics and programming benchmarks, improving six-benchmark avg@8 by 3.7-6.6 points over the strongest baselines.

ArXiv ID: 2607.22489

Authors: Jianghui Wang, Silong Yong, Francesco Orabona, Marco Canini, Katia P. Sycara, Yaqi Xie

Abstract:

Low-Rank Adaptation (LoRA) has become a widely adopted technique for efficient neural network fine-tuning, decomposing model updates into low-rank matrices. However, LoRA remains computationally costly because it updates all matrices uniformly, regardless of their actual contribution to adaptation. This cost is especially prohibitive for large-scale models with billions of parameters and for resource-constrained settings such as edge deployment and on-device fine-tuning. We show for the first time that not all LoRA matrices are equally worth tuning: matrices with smaller condition numbers (the ratio of largest to smallest singular value) are already well-balanced across directions and contribute only marginally to adaptation, whereas matrices with larger condition numbers contain underdeveloped directions that span richer subspaces and drive most of the performance gains. This observation itself is a key contribution of our work, and it motivates a more selective approach to fine-tuning. Building on this insight, we propose \k{appa}-LoRA, a method that optimizes LoRA by focusing updates on the matrices with the largest condition numbers, which capture the most informative directions of change. By restricting LoRA updates to the top 50% of weight matrices ranked by condition number, \k{appa}-LoRA halves the trainable parameter count and correspondingly reduces compute and memory cost. Extensive experiments across multiple benchmarks show that this design cuts fine-tuning time by 16.2% on average while matching the accuracy of standard LoRA and reducing memory cost by 4.5%. Further analysis reveals that the condition numbers of the selected matrices consistently decrease over training, suggesting that \k{appa}-LoRA's effectiveness stems from targeted spectral rebalancing rather than parameter selection alone.

ArXiv ID: 2607.22264

Authors: Yuxuan Liu, Joshua Placidi, Jinpei Han, Alfred John Balston, Marek Rei, A. Aldo Faisal

Abstract:

Autoregressive foundation models trained on tokenized electronic health records (EHRs) can support zero-shot clinical prediction, yet most operate on structured event codes alone, and do not incorporate multiple modalities in a principled way. We present a framework for conditioning such models on auxiliary clinical modalities, including ECG waveforms, chest X-ray images, and clinical notes, using modality-specific latent compression and gated cross-attention with temporal alignment. We investigate two key design choices: (1) how to compress long per-modality sequences (e.g., ECG time series) before they enter the multi-modal cross-attention. This feature may be essential to reduce compute overheads and may be beneficial for generalization; (2) how the choice of pretrained encoder for each modality impacts downstream performance. Through controlled ablations on MIMIC-IV, we show that the best latent-compression configurations outperforms both uncompressed cross-attention and mean pooling. Encoder choice has a clear within-modality effect, with stronger pretrained encoders consistently outperforming weaker alternatives. We further show that merely adding auxiliary modalities does not guarantee improvement on ICU mortality prediction over an EHR-only baseline. This implies that careful design of the fusion architecture and an appropriate evaluation in the clinical context are required.

ArXiv ID: 2607.21655

Authors: Jianshu Zhang, Keliang Wu, Haoran Lu, Anbang Liu, Ce Zhang, Weijie Yin, Chengxuan Qian, Xiyuan Yang, Zhenyu Pan, Guo Ye, Han Liu

Abstract:

Robotic learning takes place in dynamic environments with large behavior spaces. A terminal success signal only tells the robot whether the task is completed. It does not explain whether the current behavior is making progress, remaining unchanged, or undoing earlier progress. For this reason, recent studies have increasingly explored progress rewards that provide feedback during task execution. However, the current literature lacks a shared framework. Existing methods use different observations, goal specifications, output signals, supervision sources, and evaluation protocols. This makes it difficult to compare them and understand what their results actually validate. In this survey, we provide a unified view of progress reward modeling for robotic learning. We organize the field in three connected steps. We first study the interface of a progress model. This defines the problem from the outside by asking what information the model receives and what form of progress signal it produces. We then move inside the model and study the methods used to construct this signal. This reveals the different assumptions and mechanisms behind progress estimation and reward generation. Finally, we examine the data and benchmarks that support these methods. This shows how progress supervision is obtained and what different evaluations actually measure. Together, these three perspectives connect what a progress model is, how it is built, and how its quality is validated. We further summarize the main limitations of current approaches and discuss future research directions.

ArXiv ID: 2607.21648

Authors: Yun-Hao Tsai, Cong-Thanh Vu, Yen-Chen Liu

Abstract:

The human-like morphology of humanoid robots grants them exceptional potential for agile and versatile motor capabilities, but it also introduces significant challenges in acquiring complex skills. Traditional Learning-from-Demonstrations methods are often constrained by the high cost of collecting real-world data, the difficulty of capturing motion-specific behaviors, and the limited diversity of demonstrations across individuals. Moreover, even for the same task, humans may execute the motion in multiple distinct ways. In this paper, we propose a new framework that leverages the power of Generative AI to convert textual prompts into realistic and diverse sequences of human body movements, enabling the robot to observe multiple variations of how a single task can be performed. These synthetic demonstrations are then used as a training resource, allowing the robot to learn a broad range of task-execution styles without requiring direct human intervention. We evaluate the proposed method across four simulation scenarios. Experimental results show that the robot not only completes the tasks successfully but also demonstrates strong adaptability to complex variations in motion.

ArXiv ID: 2607.21661

Authors: Leesai Park, Jiho HOng, Sanghyun Kim

Abstract:

Diffusion policies generate multimodal robot action sequences from demonstrations, but steering them toward deployment-time constraints typically relies on differentiable guidance costs. This excludes many practical safety constraints, such as binary collision checks, joint limits, and black-box rollout costs that are nondifferentiable. We propose Gradient-free Robot Action generation via Combined diffusion-MPPI posterior mean Estimation (GRACE), which guides a pretrained diffusion policy with Model Predictive Path Integral (MPPI) control using only forward cost evaluations. Building on the common score-ascent structure of diffusion and MPPI, GRACE constructs a cost-conditioned guidance posterior at each reverse step and estimates its mean with a single MPPI update centered at the diffusion reverse mean. For differentiable costs, GRACE recovers conventional gradient guidance under a first-order, matched-covariance approximation. GRACE attains higher success rates than diffusion-based and sampling-based baselines in simulation. On a real 7-DoF manipulator, GRACE avoids a deployment-time obstacle that the unguided prior collides with in every trial. Code and experiment videos are available at https://anonymous.4open.science/w/grace-70BB/.

ArXiv ID: 2607.21670

Authors: Chaoqi Liu, Yue Zhao, Haonan Chen, Xiaoshen Han, Jiawei Gao, Ehsan Adeli, Yilun Du

Abstract:

Action tokenization maps continuous robot action chunks to discrete tokens and has become an important interface for modern visuomotor policies. Existing approaches either rely on analytical discretization methods that produce prohibitively long token sequences or learned latent tokenizers that lack structure, limiting their compatibility with downstream policies. In this work, we identify three desiderata for action tokenization - high compression, total decodability, and an ordered token space - and introduce Ordered Action Tokenization (OAT), a learned action tokenizer that satisfies all three. OAT discretizes action chunks into an ordered sequence of tokens using a transformer with registers, finite scalar quantization, and ordering-inducing training mechanisms. By training each token prefix to decode into a valid action chunk, OAT places coarse control information in early tokens and uses later tokens to refine residual detail, yielding an anytime tradeoff between inference cost and action fidelity. We validate OAT in two prevailing uses of action tokens: autoregressive policies that generate tokens for control, and token co-training policies that use token losses to shape the vision-language model context consumed by a flow-based action expert. Across three policy backbones and more than 60 tasks spanning five simulation benchmarks and real-world settings, OAT consistently delivers strong policy performance while offering significantly greater flexibility at inference time.

ArXiv ID: 2607.21918

Authors: Siqi Fan, Mingcong Chen, Ran Liu, Zixuan Yang, Xiaoyu Fu, Xiaoqing Gao, Yunhui Liu, Hongbin Liu

Abstract:

We present an action-conditioned world model framework for goal plane probe guidance in robotic ultrasound, with a focus on neck ultrasound scanning. Autonomous ultrasound tasks often require large numbers of probe-motion trajectories for training, but collecting high-quality demonstrations is labor-intensive and explicit simulators are difficult to build because ultrasound appearance depends on contact, tissue deformation, and view-dependent acoustic artifacts. We address this problem with a two-stage model-based learning pipeline. First, a latent conditional diffusion world model predicts future ultrasound observations from recent context frames, probe motions and temporal offset. Second, a goal-conditioned temporal transformer predicts ordered probe motions and is fine-tuned using rewards from the frozen world model. Experiments on the self-collected dataset show that the world model preserves action-dependent anatomical structure on target-directed scans. In real-world closed loop experiments, the framework achieves success rates of 70.0\% for carotid guidance and 65.0\% for thyroid guidance. These results demonstrate the potential of learned ultrasound dynamics for training goal-directed robotic probe navigation.