arXiv subjects

Dinesh Manocha

Publications and source records attributed to Dinesh Manocha.

At least 73 records · Page 4Linked to original sources

arXiv2026

SABER: A Stealthy Agentic Black-Box Attack Framework for Vision-Language-Action Models

Vision-language-action (VLA) models enable robots to follow natural-language instructions grounded in visual observations, but the instruction channel also introduces a critical vulnerability: small textual perturbations can alter downstream robot behavior. Systematic robustness evaluation therefore requires a black-box attacker that can generate minimal yet effective instruction edits across diverse VLA models. To this end, we present SABER, an agent-centric approach for automatically generating instruction-based adversarial attacks on VLA models under bounded edit budgets. SABER uses a GRPO-trained ReAct attacker to generate small, plausible adversarial instruction edits using character-, token-, and prompt-level tools under a bounded edit budget that induces targeted behavioral degradation, including task failure, unnecessarily long execution, and increased constraint violations. On the LIBERO benchmark across six state-of-the-art VLA models, SABER reduces task success by 20.6%, increases action-sequence length by 55%, and raises constraint violations by 33%, while requiring 21.1% fewer tool calls and 54.7% fewer character edits than strong GPT-based baselines. These results show that small, plausible instruction edits are sufficient to substantially degrade robot execution, and that an agentic black-box pipeline offers a practical, scalable, and adaptive approach for red-teaming robotic foundation models. The codebase is publicly available at https://github.com/wuxiyang1996/SABER.

cs.RO↗

arXiv2026

Do Audio-Visual Large Language Models Really See and Hear?

Audio-Visual Large Language Models (AVLLMs) are emerging as unified interfaces to multimodal perception. We present the first mechanistic interpretability study of AVLLMs, analyzing how audio and visual features evolve and fuse through different layers of an AVLLM to produce the final text outputs. We find that although AVLLMs encode rich audio semantics at intermediate layers, these capabilities largely fail to surface in the final text generation when audio conflicts with vision. Probing analyses show that useful latent audio information is present, but deeper fusion layers disproportionately privilege visual representations that tend to suppress audio cues. We further trace this imbalance to training: the AVLLM's audio behavior strongly matches its vision-language base model, indicating limited additional alignment to audio supervision. Our findings reveal a fundamental modality bias in AVLLMs and provide new mechanistic insights into how multimodal LLMs integrate audio and vision.

cs.AI↗

arXiv2026

Code Comprehension then Auditing for Unsupervised LLM Evaluation

Large Language Models (LLMs) for unsupervised code correctness evaluation have recently gained attention because they can judge if code runs as intended without requiring reference implementations or unit tests, which may be unavailable, sparse, or unreliable. However, most prior approaches condition LLM evaluators directly on the full code implementation, forcing the model to jointly infer program behavior and evaluate correctness in a single step. This entanglement leads to misinterpretations of code behavior and unreliable judgments. To mitigate this issue, we introduce CoCoA, an unsupervised Code Comprehension then Auditing framework that first comprehends functionality to generate a natural-language explanation. Then it evaluates task alignment based on this explanation. By sequentially sampling comprehension before evaluation, CoCoA improves the quality of inferred program behavior and enables the evaluator to focus on behavioral alignment rather than raw implementation details. Across multiple datasets, programming languages, and models, CoCoA achieves up to $68\%$ increased F1 score and up to $20\%$ increased accuracy over the best-performing baselines.

cs.AI↗

arXiv2026

Audio Hallucination Attacks: Probing the Reliability of Large Audio Language Models

Large Audio Language Models (LALMs) achieve strong performance on audio-language tasks; however, their reliability in real-world settings remains underexplored. We introduce Audio Hallucination Attacks (AHA), an attack suite called AHA-Eval, comprising 6.5K QA pairs designed to test whether LALMs genuinely ground their responses in the audio input. AHA targets two attack surfaces: (i) query-based attacks, which exploit question structure to induce hallucinations about absent sounds, and (ii) audio-based attacks, which inject synthetic speech describing non-existent events into the audio stream. Evaluating state-of-the-art LALMs, including Audio Flamingo 3 and Gemini 3 Pro, we observe high attack success rates of 95.35% and 79.65%, respectively, revealing a reliability gap that is hidden by standard benchmark performance. To mitigate this, we propose a 120K QA post-alignment dataset, AHA-Guard, which successfully reduces attack success rates by up to 49%.

cs.SD↗

arXiv2026

Zero-Shot Personalized Camera Motion Control for Image-to-Video Synthesis

Specifying nuanced and compelling camera motion remains a significant hurdle for non-expert creators using generative tools, creating an "expressive gap" where generic text prompts fail to capture cinematic vision. This barrier limits individual creativity and restricts the accessibility of cinematic production for small-scale industries and educational content creators. To address this, we present a zero-shot diffusion-based framework for personalized camera motion control, enabling the transfer of cinematic movements from a single reference video onto a user-provided static image without requiring 3D data, predefined trajectories, or complex graphical interfaces. Our technical contribution involves an inference-time optimization strategy using dual Low-Rank Adaptation (LoRA) networks, with an orthogonality regularizer that encourages separation between spatial appearance and temporal motion updates, alongside a homography-based refinement strategy that provides weak geometric guidance. We evaluate our approach using a new metric, CameraScore, and two distinct user studies. A 72-participant perceptual study demonstrates that our method significantly outperforms existing baselines in motion accuracy (90.45% preference) and scene preservation (70.31% preference). Furthermore, a 12-participant task-based interaction study confirms that our workflow significantly improves usability and creative control (p < 0.001) compared to standard text- or preset-based prompts. We hope this work lays a foundation for future advancements in camera motion transfer across diverse scenes.

cs.CV↗

arXiv2026

Wid3R: Wide Field-of-View 3D Reconstruction via Camera Model Conditioning

We present Wid3R, a feed-forward neural network for multi-view visual geometry reconstruction that supports wide field-of-view camera models. Unlike existing methods that assume rectified or pinhole inputs, Wid3R directly models wide-angle imagery without explicit calibration or undistortion. Our approach leverages a ray-based representation with spherical harmonics and introduces a novel camera model token to enable distortion-aware reconstruction. To the best of our knowledge, Wid3R is the first multi-frame feed-forward 3D reconstruction method that supports 360 imagery. Moreover, we show that conditioning on diverse camera types improves generalization to 360 scenes and alleviates data sparsity issues. Wid3R achieves significant performance gains, improving AUC@30 by up to +33.67 on Zip-NeRF (fisheye) and +77.33 on Stanford2D3D (360).

cs.CV↗

arXiv2026

SLAT-Phys: Fast Material Property Field Prediction from Structured 3D Latents

Estimating the material property field of 3D assets is critical for physics-based simulation, robotics, and digital twin generation. Existing vision-based approaches are either too expensive and slow or rely on 3D information. We present SLAT-Phys, an end-to-end method that predicts spatially varying material property fields of 3D assets directly from a single RGB image without explicit 3D reconstruction. Our approach leverages spatially organised latent features from a pretrained 3D asset generation model that encodes rich geometry and semantic prior, and trains a lightweight neural decoder to estimate Young's modulus, density, and Poisson's ratio. The coarse volumetric layout and semantic cues of the latent representation about object geometry and appearance enable accurate material estimation. Our experiments demonstrate that our method provides competitive accuracy in predicting continuous material parameters when compared against prior approaches, while significantly reducing computation time. In particular, SLAT-Phys requires only 9.9 seconds per object on an NVIDIA RTXA5000 GPU and avoids reconstruction and voxelization preprocessing. This results in 120x speedup compared to prior methods and enables faster material property estimation from a single image.

cs.CV↗

arXiv2026

GaussianSSC: Triplane-Guided Directional Gaussian Fields for 3D Semantic Completion

We present \emph{GaussianSSC}, a two-stage, grid-native and triplane-guided approach to semantic scene completion (SSC) that injects the benefits of Gaussians without replacing the voxel grid or maintaining a separate Gaussian set. We introduce \emph{Gaussian Anchoring}, a sub-pixel, Gaussian-weighted image aggregation over fused FPN features that tightens voxel--image alignment and improves monocular occupancy estimation. We further convert point-like voxel features into a learned per-voxel Gaussian field and refine triplane features via a triplane-aligned \emph{Gaussian--Triplane Refinement} module that combines \emph{local gathering} (target-centric) and \emph{global aggregation} (source-centric). This directional, anisotropic support captures surface tangency, scale, and occlusion-aware asymmetry while preserving the efficiency of triplane representations. On SemanticKITTI~\cite{behley2019semantickitti}, GaussianSSC improves Stage~1 occupancy by +1.0\% Recall, +2.0\% Precision, and +1.8\% IoU over state-of-the-art baselines, and improves Stage~2 semantic prediction by +1.8\% IoU and +0.8\% mIoU.

cs.RO↗

arXiv2026

Inst4DGS: Instance-Decomposed 4D Gaussian Splatting with Multi-Video Label Permutation Learning

We present Inst4DGS, an instance-decomposed 4D Gaussian Splatting (4DGS) approach with long-horizon per-Gaussian trajectories. While dynamic 4DGS has advanced rapidly, instance-decomposed 4DGS remains underexplored, largely due to the difficulty of associating inconsistent instance labels across independently segmented multi-view videos. We address this challenge by introducing per-video label-permutation latents that learn cross-video instance matches through a differentiable Sinkhorn layer, enabling direct multi-view supervision with consistent identity preservation. This explicit label alignment yields sharp decision boundaries and temporally stable identities without identity drift. To further improve efficiency, we propose instance-decomposed motion scaffolds that provide low-dimensional motion bases per object for long-horizon trajectory optimization. Experiments on Panoptic Studio and Neural3DV show that Inst4DGS jointly supports tracking and instance decomposition while achieving state-of-the-art rendering and segmentation quality. On the Panoptic Studio dataset, Inst4DGS improves PSNR from 26.10 to 28.36, and instance mIoU from 0.6310 to 0.9129, over the strongest baseline.

cs.CV↗

arXiv2026

TransCurriculum: Multi-Dimensional Curriculum Learning for Fast & Stable Locomotion

High-speed legged locomotion struggles with stability and transfer losses at higher command velocities during deployment. One reason is that most curricula vary difficulty along single axis, for example increase the range of command velocities, terrain difficulty, or domain parameters (e.g. friction or payload mass) using either fixed update rule or instantaneous rewards while ignoring how the history of robot training has evolved. We propose TransCurriculum, a transformer-based multi-dimensional curriculum learning approach for agile quadrupedal locomotion. TransCurriculum adapts to 3 axes, velocity command targets, terrain difficulty, and domain randomization parameters (friction and payload mass). Rather than feeding task reward history directly into the low-level control policy, our formulation exploits it at the curriculum level. A transformer-based teacher retrieves the sequence of rewards and uses it to predict future rewards, success rate, and learning progress to guide expansion of this multidimensional curriculum towards high performing task bins. Finally we validate our approach on the Unitree Go1 robot in simulation (Isaac Gym) and deploy it zero-shot on Go1 hardware. Our TransCurriculum policy achieves a maximum velocity of 6.3 m/s in simulation and outperforms prior curriculum baselines. We tested our TransCurriculum trained policy on terrains (carpets, slopes, tiles, concrete), achieving a forward velocity of 4.1 m/s on carpet surpassing the fastest curriculum methods by 18.8% and achieves maximum zero-shot value among all tested methods. Our multi-dimensional curriculum also reduces the transfer loss to 18% from 27% for command only curriculum, demonstrating the benefits of joint training over velocity, terrain and domain randomization dimension while keeping the task success rate of 80-90% on rigid indoor and outdoor surfaces.

cs.RO↗

arXiv2026

Multi-Domain Audio Question Answering Benchmark Toward Acoustic Content Reasoning

We present Task 5 of the DCASE 2025 Challenge: an Audio Question Answering (AQA) benchmark spanning multiple domains of sound understanding. This task defines three QA subsets (Bioacoustics, Temporal Soundscapes, and Complex QA) to test audio-language models on interactive question-answering over diverse acoustic scenes. We describe the dataset composition (from marine mammal calls to soundscapes and complex real-world clips), the evaluation protocol (top-1 accuracy with answer-shuffling robustness), and baseline systems (Qwen2-Audio-7B, AudioFlamingo 2, Gemini-2-Flash). Preliminary results on the development set are compared, showing strong variation across models and subsets. This challenge aims to advance the audio understanding and reasoning capabilities of audio-language models toward human-level acuity, which are crucial for enabling AI agents to perceive and interact about the world effectively.

cs.SD↗

arXiv2026

FALCON: Future-Aware Learning with Contextual Object-Centric Pretraining for UAV Action Recognition

We introduce FALCON, a unified self-supervised video pretraining approach for UAV action recognition from raw RGB aerial footage, requiring no additional preprocessing at inference. UAV videos exhibit severe spatial imbalance: large, cluttered backgrounds dominate the field of view, causing reconstruction-based pretraining to waste capacity on uninformative regions and under-learn action-relevant human/object cues. FALCON addresses this by integrating object-aware masked autoencoding with object-centric dual-horizon future reconstruction. Using detections only during pretraining, we construct objectness priors that (i) enforce balanced token visibility during masking and (ii) concentrate reconstruction supervision on action-relevant regions, preventing learning from being dominated by background appearance. To promote temporal dynamics learning, we further reconstruct short- and long-horizon future content within an object-centric supervision region, injecting anticipatory temporal supervision that is robust to noisy aerial context. Across UAV benchmarks, FALCON improves top-1 accuracy by 2.9\% on NEC-Drone and 5.8\% on UAV-Human with a ViT-B backbone, while achieving 2$\times$--5$\times$ faster inference than supervised approaches that rely on heavy test-time augmentation.

cs.CV↗

arXiv2026

ViLAM: Distilling Vision-Language Reasoning into Attention Maps for Social Robot Navigation

We introduce ViLAM, a novel method for distilling vision-language reasoning from large Vision-Language Models (VLMs) into spatial attention maps for socially compliant robot navigation. Unlike traditional methods that rely on expert demonstrations or human-annotated datasets, ViLAM performs knowledge distillation and fine-tuning at the intermediate layer representation (attention) level by aligning attention maps from a pretrained vision-action model with socially guided attention maps derived from a large VLM. These distilled attention maps highlight key navigational regions in a scene and serve as socially informed spatial cost maps for motion planning. To achieve this, we introduce a novel attention-level distillation loss that fuses knowledge from both sources, generating augmented attention maps with enhanced social awareness. These refined attention maps are then used as a traversability costmap within a socially aware local planner for navigation. We validate our approach through real-world experiments on a Husky wheeled robot, and demonstrate 14.2% - 50% improvements in success rate over existing methods.

cs.RO↗

arXiv2026

ROVER: Regulator-Driven Robust Temporal Verification of Black-Box Robot Policies

We present a novel, regulator-driven approach for the temporal verification of black-box autonomous robot policies, inspired by real-world certification processes where regulators often evaluate observable behavior without access to model internals. Central to our method is a regulator-in-the-loop approach that evaluates execution traces from black-box policies against temporal safety requirements. These requirements, expressed as prioritized Signal Temporal Logic (STL) specifications, characterize behavior changes over time and encode domain knowledge into the verification process. We use Total Robustness Value (TRV) and Largest Robustness Value (LRV) to quantify average performance and worst-case adherence, and introduce Average Violation Robustness Value (AVRV) to measure average specification violation. Together, these metrics guide targeted retraining and iterative model improvement. Our approach accommodates diverse temporal safety requirements (e.g., lane-keeping, delayed acceleration, and turn smoothness), capturing persistence, sequencing, and response across two distinct domains (virtual racing game and mobile robot navigation). Across six STL specifications in both scenarios, regulator-guided retraining increased satisfaction rates by an average of 43.8%, with consistent improvement in average performance (TRV) and reduced violation severity (LRV) in half of the specifications. Finally, real-world validation on a TurtleBot3 robot demonstrates a 27% improvement in smooth-navigation satisfaction, yielding smoother paths and stronger compliance with STL-defined temporal safety requirements.

cs.RO↗

arXiv2026

InfinityStory: Unlimited Video Generation with World Consistency and Character-Aware Shot Transitions

Generating long-form storytelling videos with consistent visual narratives remains a significant challenge in video synthesis. We present a novel framework, dataset, and a model that address three critical limitations: background consistency across shots, seamless multi-subject shot-to-shot transitions, and scalability to hour-long narratives. Our approach introduces a background-consistent generation pipeline that maintains visual coherence across scenes while preserving character identity and spatial relationships. We further propose a transition-aware video synthesis module that generates smooth shot transitions for complex scenarios involving multiple subjects entering or exiting frames, going beyond the single-subject limitations of prior work. To support this, we contribute with a synthetic dataset of 10,000 multi-subject transition sequences covering underrepresented dynamic scene compositions. On VBench, InfinityStory achieves the highest Background Consistency (88.94), highest Subject Consistency (82.11), and the best overall average rank (2.80), showing improved stability, smoother transitions, and better temporal coherence.

cs.CV↗

arXiv2026

CHOP: Counterfactual Human Preference Labels Improve Obstacle Avoidance in Visuomotor Navigation Policies

Visuomotor navigation policies have shown strong perception-action coupling for embodied agents, yet they often struggle with safe navigation and dynamic obstacle avoidance in complex real-world environments. We introduce CHOP, a novel approach that leverages Counterfactual Human Preference Labels to align visuomotor navigation policies towards human intuition of safety and obstacle avoidance in navigation. In CHOP, for each visual observation, the robot's executed trajectory is included among a set of counterfactual navigation trajectories: alternative trajectories the robot could have followed under identical conditions. Human annotators provide pairwise preference labels over these trajectories based on anticipated outcomes such as collision risk and path efficiency. These aggregated preferences are then used to fine-tune visuomotor navigation policies, aligning their behavior with human preferences in navigation. Experiments on the SCAND dataset show that visuomotor navigation policies fine-tuned with CHOP reduce near-collision events by 49.7%, decrease deviation from human-preferred trajectories by 45.0%, and increase average obstacle clearance by 19.8% on average across multiple state-of-the-art models, compared to their pretrained baselines. These improvements transfer to real-world deployments on a Ghost Robotics Vision60 quadruped, where CHOP-aligned policies improve average goal success rates by 24.4%, increase minimum obstacle clearance by 6.8%, reduce collision and intervention events by 45.7%, and improve normalized path completion by 38.6% on average across navigation scenarios, compared to their pretrained baselines. Our results highlight the value of counterfactual preference supervision in bridging the gap between large-scale visuomotor policies and human-aligned, safety-aware embodied navigation.

cs.RO↗

arXiv2026

MotionHint: Self-Supervised Monocular Visual Odometry with Motion Constraints

We present a novel self-supervised algorithm named MotionHint for monocular visual odometry (VO) that takes motion constraints into account. A key aspect of our approach is to use an appropriate motion model that can help existing self-supervised monocular VO (SSM-VO) algorithms to overcome issues related to the local minima within their self-supervised loss functions. The motion model is expressed with a neural network named PPnet. It is trained to coarsely predict the next pose of the camera and the uncertainty of this prediction. Our self-supervised approach combines the original loss and the motion loss, which is the weighted difference between the prediction and the generated ego-motion. Taking two existing SSM-VO systems as our baseline, we evaluate our MotionHint algorithm on the standard KITTI benchmark. Experimental results show that our MotionHint algorithm can be easily applied to existing open-sourced state-of-the-art SSM-VO systems to greatly improve the performance by reducing the resulting ATE by up to 28.73%.

cs.CV↗

arXiv2026

LoLep: Single-View View Synthesis with Locally-Learned Planes and Self-Attention Occlusion Inference

We propose a novel method, LoLep, which regresses Locally-Learned planes from a single RGB image to represent scenes accurately, thus generating better novel views. Without the depth information, regressing appropriate plane locations is a challenging problem. To solve this issue, we pre-partition the disparity space into bins and design a disparity sampler to regress local offsets for multiple planes in each bin. However, only using such a sampler makes the network not convergent; we further propose two optimizing strategies that combine with different disparity distributions of datasets and propose an occlusion-aware reprojection loss as a simple yet effective geometric supervision technique. We also introduce a self-attention mechanism to improve occlusion inference and present a Block-Sampling Self-Attention (BS-SA) module to address the problem of applying self-attention to large feature maps. We demonstrate the effectiveness of our approach and generate state-of-the-art results on different datasets. Compared to MINE, our approach has an LPIPS reduction of 4.8%-9.0% and an RV reduction of 73.9%-83.5%. We also evaluate the performance on real-world images and demonstrate the benefits.

cs.CV↗