Computer Science (arXiv)

Existing visual navigation models often aim to develop foundation models that can generalize robot navigation across diverse platforms. However, many of these models are prone to collisions when deployed in complex indoor environments, particularly in structured layouts and narrow passages. To address this problem, we propose a depth image- and point-goal-conditioned navigation system, DevGRU. The proposed system employs an action predictor (AP) that generates collision-aware future trajectories, enabling effective avoidance of immediate obstacles. In conjunction with a collision predictor, the AP further compensates for errors accumulated in the goal pose estimation and proactively mitigates future deviations. To evaluate our method, we conducted experiments across nine different scenes and three state-of-the-art approaches - ViNT, NoMaD, and NavDP - as well as four additional variants of ViNT and NoMaD. In terms of navigation performance, DevGRU significantly outperforms ViNT and NoMaD by a large margin. In addition, the proposed model has a relatively small number of trainable parameters, resulting in the fastest inference time among the baselines, particularly outperforming NavDP by 7x in model size and 17x in inference time.

Lightweight proxy models enable rapid experimentation without repeatedly training frontier-scale systems, but their small kernels often leave modern accelerators underutilized. Conventional training compounds this inefficiency by scheduling the forward and backward passes as disjoint phases, so spare capacity in one cannot be filled by work from the other. We reinterpret the detachment mechanism of Forward-Forward (FF) as a scheduling primitive: given a local objective, detaching a block's output removes downstream gradient dependencies, making its backward pass ready when its forward pass finishes. ERASE launches each detached subgraph's backward pass early on a separate CUDA stream, overlapping it with subsequent forward work. Execution trace on a lightweight transformer demonstrates this overlap and its limit: a kernel that saturates the device leaves no capacity for concurrency. On a large-scale click-through-rate model, detaching six dense subarchitectures improves training throughput by up to $9.51\%$ while keeping normalized entropy close to the baseline.

Proper use of the aircraft maintenance manual is essential for correct maintenance, providing procedures, diagrams, cautions, and specifications. However, technicians often avoid consulting it because it is difficult to navigate and time-consuming under strict schedules. Retrieval augmented generation (RAG) models have recently been introduced in aircraft maintenance, yet existing models focus solely on textual retrieval. This research therefore targeted the Cessna 172 Maintenance Manual (C172-MM), widely used in general aviation, and developed a multimodal manual retriever (MMR) capable of retrieving multimodal manual pages. Retrieval performance was evaluated using synthetic queries covering procedures, diagrams, caution/safety information, and specifications; the MMR achieved 93.37% recall@5. Beyond retrieval, a multimodal RAG (MRAG) pipeline was examined, in which retrieved pages were input to a vision-language model that generated responses to the synthetic queries, achieving 87.20% semantic similarity to ground-truth answers. Three practical feasibilities were also assessed: inference time, operational cost, and interpretability. Average retrieval time for five pages was 11.93 seconds and response generation took 4.95 seconds, at $0.0091 per query, while interpretability was validated through heatmap visualizations. These results indicate that the MRAG pipeline for the C172-MM can reduce the time technicians spend searching manuals and retrieving multimodal information.

The high photon flux at synchrotron beamlines allows for the measurement of fast dynamical processes. However, beamline radiation-safety protocols prohibit human intervention during X-ray experiments, limiting the ability to perform versatile real-time sample manipulations during continuous data acquisition. Here we present a robotic platform at an X-ray scattering beamline to enable real-time sample handling and processing in the experimental hutch, revealing previously inaccessible transient in-situ dynamics in perovskite thin films. This modular multi-tool robotic architecture enables in-hutch sample manipulation beyond human-access constraints, establishing a foundation for automated and autonomous synchrotron experimentation.

This paper investigates the sensing-centric design of reconfigurable intelligent surface (RIS)-enabled rate-splitting multiple access-integrated sensing and communication (RSMA-ISAC) systems. Specifically, we propose a new beam-gain approximation method to enhance the sensing beam gain while satisfying communication quality-of-service (QoS) constraints.Since the joint optimization of the beamforming vectors and RIS phase shifts is highly coupled and non-convex, existing methods typically rely on generic optimization solvers involving substantial computational complexity. To address this issue, we propose an efficient constraints-separation-based alternating optimization algorithm (CS-AO). Our proposed algorithm effectively decouples the optimization variables and yields closed-form solutions for all subproblems, thereby significantly reducing the computational burden. Simulation results show that the proposed algorithm achieves sensing beam-gain performance comparable to successive convex approximation (SCA) and semidefinite relaxation (SDR) benchmarks, while achieving more than 120-fold and 50-fold runtime reductions. In addition, compared with conventional space-division multiple access (SDMA) schemes, the proposed design exhibits substantial sensing beam gain.

Robotic assembly layout planning must determine the assembly site and the initial pose of each part while ensuring collision-free execution of a prescribed assembly sequence. This problem is challenging because the obstacle environment changes after each assembly step, and unassembled parts re maining in the workspace may block robot motions. We observe that the feasibility of each assembly step depends only on the initial poses of the current and later-assembled parts. Based on this dependency, we propose Backward Layout Search (BLS), which assigns initial part poses in reverse assembly order. Each expansion performs geometric, kinematic, grasp, and prescribed-motion checks, while collision masks and candidate set filtering remove infeasible initial part pose candidates. Promising partial layouts are retained through beam selection, and complete layouts are validated by full motion planning in forward assembly order. Experiments on five assembly models show that BLS produces collision-free executable layouts and reduces step evaluations and search time compared with a matched forward search.

\textbf{Background and Objective}: Reliable atrial fibrillation (AF) detection from electrocardiogram (ECG) signals remains challenging in real-world clinical settings due to variable lead configurations, cross-dataset domain shifts, and pervasive physiological and technical artifacts. So we develop a robust and generalizable deep learning model for accurate AF detection.\\ \textbf{Methods}: We propose the Dual-Codebook Graph Collaborative Network (DCGCNet), a novel end-to-end vector-quantized variational autoencoder that jointly performs AF classification and ECG reconstruction. DCGCNet introduces two key components: (1) a Local-Global Contrastive Module for learning noise-invariant representations, and (2) an Adaptive Codebook Vector Quantizer that dynamically refines codebook prototypes to better align with input data distributions, thereby preventing codebook collapse and enhancing generalization.\\ \textbf{Results}: DCGCNet achieves state-of-the-art performance in standard intra-dataset 12-lead evaluation and demonstrates exceptional cross-dataset generalization across seven diverse settings, consistently attaining AUC > 0.98 in all cases. Furthermore, it maintains high diagnostic accuracy under realistic noisy conditions, including baseline wander, powerline interference, and EMG artifacts.\\ \textbf{Conclusions}: DCGCNet establishes a new benchmark for robust, generalizable, and noise-resilient AF detection, showing strong potential for deployment in real-world clinical environments.

Falls among older adults represent a major public health challenge driven by complex, time-varying interactions across multiple risk domains. Effective fall risk factor identification requires learning from heterogeneous longitudinal data while accounting for sparse and delayed fall-related outcome events. However, existing approaches are largely static and fail to adaptively model evolving, individualized risk factors across modalities and time. We propose PAFIR, a Personalized and Adaptive Feature selection framework for fall risk Identification and pRevention, which formulates adaptive feature selection as a reinforcement learning problem over longitudinal multimodal health data. PAFIR jointly models structural dependencies among correlated assessment variables and temporal dynamics in wearable-derived physical activity data, and learns adaptive selection policies across repeated study visits using reward signals derived from sparse fall incidence outcomes. We apply PAFIR to data from the Physio fEedback Exercise pRogram (PEER) cluster-randomized trial. Experimental results demonstrate that PAFIR more effectively captures longitudinal and structural patterns of feature relevance than state-of-the-art baselines, and enables dynamic, subject-specific feature selection. By adapting selected features over time, PAFIR supports more timely and personalized fall prevention strategies.

More context does not require a larger retrieval budget. Under the same ceiling, a retrieval system can recover more of the evidence a question requires by following relationships between evidence that flat top-k ranking leaves behind. We test that proposition with Dual-Bounded Relational Recall (DBRR), which allocates a fixed retrieval budget between relevance-selected seeds and bounded graph-adjacent context, against matched flat top-k retrieval using the same relevance-ranking stage and the same maximum number of retrieval units and tokens. The outcome is complete recovery of the official HotpotQA supporting-evidence set for each question. Across 7,405 FullWiki questions, the Primary DBRR allocation increased complete supporting-evidence recovery by 23.8 percentage points over its matched flat baseline (paired risk difference 0.2377; question-level bootstrap 95% interval 0.2269 to 0.2489). It improved 1,952 questions, tied on 5,261, and harmed 192. Bridge questions drove the effect, with a 28.7-point increase; comparison questions showed a smaller 4.2-point difference. In a prespecified, evaluation-only diagnostic population, real relationships also outperformed random-neighbor and degree-preserving shuffled-graph controls. The result is straightforward: under the same context budget, complete-evidence retrieval depends not only on which items rank highest, but on how context is allocated around them. Relational allocation recovered complete evidence sets that flat top-k retrieval left incomplete.

End-to-end imitation learning avoids hand-made robot motion for approaching and grasping, but the policy must still decide which fruit to pick and where to close the gripper. Occlusion can make the policy lose the selected fruit during harvesting, and the correct closing point is difficult to infer from pixels alone. This paper presents HarvestPoint-ACT, which makes both decisions explicit in perception and provides them to the policy. An instance segmentation front end with a keypoint branch predicts a mask and a harvest point for each visible fruit, where the harvest point specifies the location to close the gripper. A scheduler ranks detected candidates by occlusion and travel distance and selects one target. After each attempt, it redetects and reranks the candidates because the canopy may have changed. The selected fruit is encoded for an action chunking transformer as an eight-dimensional state, containing the absolute harvest point, the vector from the gripper to that point, a validity flag, and a confidence score. When the selected fruit is temporarily undetected, the system retains the last harvest point estimate in the robot base frame and marks it as stale, and aborts the attempt if the loss persists. On a canopy mock-up, HarvestPoint-ACT achieves a success rate of 88%, and of 75% under heavy occlusion.

We present the first mechanised formalisation of Romanov's Triplet Logic (TLS) in the Rocq proof assistant. TLS is a triplet-based combinatorial framework for reasoning about compatible paths through layered triplet structures, called Compact Triplets Structures (CTS), and their intersection via Romanov's Effective Procedure, which we refer to as Simple Vertex Intersection (SVI). Originally motivated by Boolean satisfiability, TLS constitutes a self-contained mathematical theory whose formal properties had not been previously established. We formalise the core of TLS in Rocq, including Compact Triplets Formulas (CTF), CTS, hyperstructures, clearing, and SVI. For the well-formed sliding-window fragment we verify a clause-by-clause CNF-to-CTF translation, the clearing procedure, and aligned intersection, and we prove explicit polynomial-time bounds for the filter stages. Our main contribution is a precise correctness boundary: the existence of a joint satisfying set implies non-emptiness of SVI, but the converse does not hold in general; for aligned structures we recover a complete bi-implication, extended to systems of structures. We also formalise soundness of grouped-window translation and exhibit a formal counterexample to its completeness. We introduce VFR, an extracted OCaml prototype that provides a verified decision procedure for the sliding-window fragment and a sound one-sided filter for general 3-CNF, with a Python runtime and reproducible Docker packaging. Benchmarks on random and structured instances confirm the predicted behaviour, and the complete toolchain is available as a curated Zenodo artifact. The Rocq development comprises more than 23,000 lines of code across seventeen files, with 427 proved lemmas and theorems and zero admitted goals.

Modern mental healthcare faces a critical shortage of senior supervisory oversight, leading to a "supervision gap" where novice therapists manage high-stakes risks with delayed professional feedback. This paper proposes a new framework utilizing a fine-tuned Mistral-7B-instruct model as an automated "Supervisor-in-the-Loop" system. By leveraging 106 sessions from the DAIC-WOZ dataset, the model performs a tri-stream analysis: (1) Therapeutic Alliance tracking via semantic adherence, (2) Latent risk prediction using attention-weighted analytics, and (3) Supervisory Triage via a Dynamic Clinical Urgency Index (D-CUI). Our multi-modal VAL (Visual-Acoustic-Linguistic) framework achieves 95% technique identification accuracy [95% CI: 75.1%-99.9%], alliance assessment MAE of 0.105 on a 5-point scale [95% CI: 0.059-0.151], therapeutic fidelity alpha = 0.423, and mean D-CUI of 0.370 [95% CI: 0.322-0.419]. Training converged in 105 steps with 85.2% loss reduction on a single Tesla T4 GPU. The system reduces supervisory triage latency from 72 hours to real time (~10 seconds per session), enabling proactive intervention in high-risk cases. The system addresses the cold-start problem through Bayesian priors and implements timestamp-based modality synchronization for robust multi-modal fusion.

Tangut is an extinct language whose script does not explicitly mark word boundaries. We present the first systematic study of Tangut word segmentation using 2,750 expert-annotated segments(31,893 tokens), traditional lexicons, and unlabeled text. Our framework combines a reliability-calibrated lexicon-lattice representation, explicit distributional statistics, and a lightweight character encoder pretrained with MLM. Segment-level five-fold cross-validation shows that lexical and statistical features raise CRF F1 to approximately 0.91. The full TangutEncoder reaches the highest mean F1 (0.911) and improves recall beyond the labeled training vocabulary. These results demonstrate generalization beyond the limited supervised vocabulary across thematically diverse held-out passages, while document-level transfer remains to be evaluated.

Robot foundation models (RFMs), including vision-language-action (VLA) policies, are often discussed through a scaling view: more data, larger models, and broader benchmarks should improve generalization. In robotics, however, a model can generalize while work still remains before it can run on a robot with a particular body. The work required differs across methods and target robots, and those differences affect practical deployment. We call the gap between reusable models, representations, or data and their use in execution on the target robot the embodiment gap. This survey examines what can be reused across robot embodiments and what must still be implemented on a new robot. We place existing methods on a two-axis map that shows the type of shared structure and the stage at which adaptation is needed for execution on the target robot. We then examine recent work through three overlapping research directions: sharing semantics and perception, sharing robot data and interfaces, and learning correspondence across embodiments. We also propose a reporting framework for adaptation work that success rate alone does not reveal. The framework identifies the work that should be checked when comparing cross-embodiment learning and highlights work that remains on a new robot and questions for future study.

This paper develops a computational framework to optimize the process parameters such that the bond quality between extruded polymer filaments is maximized in fused filament fabrication (FFF). A transient heat transfer analysis providing an estimate of the temperature profile of the filaments is coupled with a sintering neck growth model to assess the bond quality that occurs at the interfaces between adjacent filaments. Predicting the variability in the FFF process is essential for achieving proactive quality control of the manufactured part; however, the models used to predict the variability are affected by assumptions and approximations. This paper systematically quantifies the uncertainty in the bond quality model prediction due to various sources of uncertainty, both aleatory and epistemic, and includes the uncertainty and the model discrepancy in the process parameter optimization. Variance-based sensitivity analysis based on Sobol indices is used to quantify the relative contributions of the different uncertainty sources to the uncertainty in the bond quality. A Gaussian process (GP) surrogate model is constructed to compute and include the model discrepancy within the optimization. Physical experiments are conducted for calibration and validation of the physics model and also for validation of the optimum solution. The results show that the proposed formulation for process parameter optimization under uncertainty results in high bond quality between adjoining filaments of the FFF product.

This paper proposes a multi-level Bayesian calibration approach that fuses information from heterogeneous sources and accounts for uncertainties in modeling and measurements for time-dependent multi-component systems. The developed methodology has two elements: quantifying the uncertainty at component and system levels, by fusing all available information, and corrected model prediction. A multi-level Bayesian calibration approach is developed to estimate component-level and system-level parameters using measurement data that are obtained at different time instances for different system components. Such heterogeneous data are consumed in a sequential manner, and an iterative strategy is developed to calibrate the parameters at the two levels. This calibration strategy is implemented for two scenarios: offline and online. The offline calibration uses data that is collected over all the time-steps, whereas online calibration is performed in real-time as new measurements are obtained at each time-step. Analysis models and observation data for the thermo-mechanical behavior of gas turbine engine rotor blades are used to analyze the effectiveness of the proposed approach.

This work presents a data-driven methodology for multi-objective optimization under uncertainty of process parameters in the fused filament fabrication (FFF) process. The proposed approach optimizes the process parameters with the objectives of minimizing the geometric inaccuracy and maximizing the filament bond quality of the manufactured part. First, experiments are conducted to collect data pertaining to the part quality. Then, Bayesian neural network (BNN) models are constructed to predict the geometric inaccuracy and bond quality as functions of the process parameters. The BNN model captures the model uncertainty caused by the lack of knowledge about model parameters (neuron weights) and the input variability due to the intrinsic randomness in the input parameters. Using the stochastic predictions from these models, different robustness-based design optimization formulations are investigated, wherein process parameters such as nozzle temperature, nozzle speed, and layer thickness are optimized under uncertainty for different multi-objective scenarios. Epistemic uncertainty in the prediction model and the aleatory uncertainty in the input is considered in the optimization. Finally, Pareto surfaces are constructed to estimate the tradeoffs between the objectives. Both the BNN models and the effectiveness of the proposed optimization methodology are validated using the actual manufacturing of the parts.

Language model agents now execute bounded tasks reliably. Whether they can sustain effective decision-making over long horizons, where actions have cumulative consequences and the environment responds to their choices, remains largely unmeasured. FM-Bench (Football Management Benchmark) measures this. An LLM agent runs a football club for 20 in-game years through 26 tools and roughly 340 to 400 decision stops. It drafts a squad on the same budget as every rival, trades players, negotiates contracts, invests in facilities and youth, sets lineups, and answers to a board that can fire it, while a deterministic engine accumulates every year into one final score with no LLM judge or human rater. The solo track plays each of 15 frontier models against a frozen scripted world, and the Arena places the same models plus a scripted anchor in one shared 20-year world; to our knowledge, the first head-to-head evaluation at this scale. We measure six behavioral capabilities behind the score. Across three seeds, all 15 models complete every horizon while the blind scripted baselines die out in most of theirs, and claude-fable-5 tops the solo board on mean score and the Arena, where the title nonetheless rotates among ten models. Neither scale, price, nor vendor predicts the order; the order settles only late in the horizon, and the best first-play human lands only at the bottom of the model board. What separates the models is managerial behavior rather than computation. Higher-scoring models reduce slow-payoff investment near the end, keep cash invested rather than idle, and open renewals well before the deadline, while token spend predicts nothing. No model learns the market's hidden prices from hundreds of rejected bids, and self-managed memory fails in two opposite modes: an archive that only grows or a plan rewritten every season. Code is available at https://github.com/Analogy-AI/fm-bench.

Online discussions of political issues do not always happen in places explicitly dedicated to political talk; they also arise in online spaces focused on at least nominally apolitical interests, identities, and/or places. Whatever one's normative view of politics entering these ``online third spaces,'' understanding who brings political issues into them, and when, requires studying these spaces at scale. In turn, studying these spaces at scale requires a construct that captures both who is speaking and who is listening, and that holds up over time. Building on Bruns' distinction between participant-centered personal publics and post-centered issue publics, we introduce the longitudinal relational networked public (or, simply, the longitudinal public): the coupling of discourse produced by a socially connected set of creators with the durable attention their shared audience gives it. The longitudinal public departs from related relational constructs in three ways: it is anchored in the attention patterns of a non-elite, population-level audience; it treats within-public structure as an object of analysis rather than assuming homogeneity; and it incorporates the audience as a force that shapes creator discourse. In a case study, we identify 150 longitudinal publics from the following ties of a panel of Twitter/X users matched to U.S. voter records, then measure their discursive overlap with the electoral politics and Black Lives Matter issue publics across 2020, a period spanning the murder of George Floyd and the general election. We find that the spaces that best fit the idea of a third space have the most politically heterogeneous audiences, and thus the most potential room for cross-partisan talk. This result, among several others, shows how a relational, audience-aware construct can reveal where and through whom political talk enters everyday online life.

Recent work has shown that large language models (LLMs) exhibit strong numerical sequence modeling capabilities and show promise in time-series prediction. While LLMs display in-context learning capabilities, the mechanisms with which they accomplish time-series prediction remain unclear. Specifically, whether they truly understand the underlying structure, which at a minimum requires reasoning over first differences in the sequence of numbers. To study this, we investigate Llama 3.1-8B from a mechanistic interpretability point of view. Mechanistic interpretability is an emerging field concerned with the reverse engineering of the algorithms learned by neural networks such as LLMs. To assess Llamas' numerical sequence modeling capabilities and to facilitate our mechanistic interpretability analysis, we create a sequence modeling task that cannot be solved without picking up structural cues. Specifically, we sample n random numbers and repeat them with an offset. We find that Llama displays strong performance on our tasks suggesting that it can pick up on the underlying structure. To understand the mechanisms that allow it to do so, we perform probing experiments and activation patching based counterfactual analysis. Probing reveals that the model computes and stores first differences in its internal representations without explicit supervision, indicating that it tracks structural information about the sequence. Activation patching reveals that Llama retrieves the relevant first-difference with a mechanism similar to an induction circuit and subsequently adds it to the current value. Notably, our work represents one of the first studies to identify this form of concept induction in LLMs.

The edge of stability refers to a phenomenon in deep learning with gradient-based optimizers where the Hessian eigenvalues of the loss remain stable above a threshold that the classical descent lemma predicts to be unstable. Previous works formulate the edge of stability with respect to the maximum Hessian eigenvalue and the learning rate. However, we observe that many first-order methods, including gradient descent, significantly violate the stability bound predicted by these theories by a factor as large as $\times 21.1$. Moreover, this deviation turns out to be systematic and highly dependent on the underlying optimizer, which is not captured by previous formulations. This calls for a new formulation of the stability threshold, which we derive from the directional Hessian and the gradient-alignment score with respect to the actual update taken by the optimizer, rather than the maximum curvature mode. Our new formulation of the realized edge of stability not only removes optimizer-dependent offsets and provides more consistent predictions of the stability threshold, but also introduces new diagnostic tools that reveal the unique role of the optimizer in actively balancing between the temporal and spatial budgets in first-order optimization.

3D Gaussian Splatting enables efficient novel view synthesis, but accurate mesh reconstruction remains difficult in weakly observed and occluded regions, where Gaussian primitives may grow into unstable or geometrically inconsistent structures. We propose CoMVS-GS, a general surface reconstruction framework that combines Multi-View Stereo with Gaussian splatting. CoMVS-GS initializes Gaussian primitives from dense multi-view stereo points with pre-flattened scales and normal-aligned orientations, providing stronger geometric priors than sparse structure-from-motion initialization and reducing ambiguity during early optimization. It further introduces PatchMatch-3DGS Mutual Supervision, where Gaussian-rendered depths and normals initialize PatchMatch refinement, and refined PatchMatch depths supervise Gaussian optimization to improve weakly constrained geometry. For surface extraction, CoMVS-GS replaces truncated signed distance field voxel fusion with a Delaunay graph-cut meshing pipeline, reducing sensitivity to voxel resolution while preserving visibility-consistent surface evidence. Experiments on DTU, GauU-Scene V2, and MatrixCity show that CoMVS-GS remains competitive on object-level reconstruction and improves geometric accuracy and mesh compactness in outdoor scenes while maintaining high rendering quality.

Effective sign language (SL) acquisition is crucial for deaf children, yet 95% are born to hearing parents who often lack proficiency in SL. SL recognition can power learning tools to help parents communicate with their children. However, Japanese Sign Language (JSL) lacks large-scale, multi-signer datasets, hindering the development of models that can generalize to new users. To address this gap, we introduce JSL-DC, the largest JSL dataset by video count, comprising 36.7K videos from 19 signers. The entire process was Deaf-centric: the lexicon comprising 270 JSL words was selected by Deaf and Coda linguists to facilitate parent-child communication, all participants were Deaf individuals who use JSL daily, and the data underwent a two-stage review process involving Deaf linguists. Moreover, we provide linguist-derived descriptions for distinguishing confusable signs. We demonstrate that the proposed model inspired by the descriptions outperforms state-of-the-art recognition methods by 9.8% on the confusable subset. The dataset, along with its linguistic description that inspires new models, will be released under a CC-BY 4.0 license to accelerate research in SL recognition.

Vision-Language-Action (VLA) models process long multimodal token sequences, making inference expensive in both memory and computation. Existing efficiency methods mainly reduce visual tokens, but aggressive token pruning becomes fragile because removing a token discards its entire representation. Sub-token compression provides a complementary alternative by retaining more tokens while reducing their value width. However, directly applying sub-token compression to VLA policies is less effective because information important for perception, language understanding, and control is distributed differently across the multimodal representation. We introduce Role-Conditioned Sub-Token Routing (RoleSub), which learns how to compress the value representations of retained tokens. After visual token reduction, RoleSub partitions each retained value representation into groups in an orthogonal space and uses a lightweight router to determine which groups should be preserved. The routing decision is conditioned on the token representation, a learned latent role representation, and language context. The same mechanism can also be applied to language values, allowing visual and language representations to be compressed without removing additional tokens. We evaluate RoleSub on OpenVLA-OFT-7B across the four LIBERO suites. At matched visual-KV budgets, RoleSub outperforms a trained token-only control in 33 of 36 settings, with the largest gains under aggressive compression. Combining visual and language compression reduces total KV to 9.2--11.3% of the original while retaining strong control performance on most tasks. These results show that reducing the representation within retained tokens provides an effective complement to token pruning for aggressive VLA compression.

Combinatorial scheduling poses a significant challenge for language models, requiring them to identify feasible solutions within exponentially large search spaces while satisfying complex constraints. This challenge is especially pronounced in resource-constrained settings, where larger language models are impractical and selection is limited to smaller models which often fail to preserve feasibility when scheduling directly from natural language. To address these limitations, we introduce SDDL, a neuro-symbolic framework that translates natural-language scheduling problems into compact, solver-aligned representations of tasks, resources, constraints, and objectives, while delegating low-level modeling and search to a deterministic compiler and external solver. On a 300-instance, multi-family subset of scheduling problems, SDDL improves independently verified feasibility for every resource-constrained model tested. The two strongest SDDL configurations reach 55.3% and 28.3%, up from direct-generation baselines of 23.7% and 1.3% and solver-code baselines of 21.7% and 7.0%, with a 0.0% median optimality gap among feasible schedules. By expressing problem structure rather than generating solutions or solver code, SDDL enables smaller models to approach the strongest evaluated direct- and solver-code configurations, including substantially larger frontier models.