Autonomous wheel-loader control requires joint reasoning over task semantics, egocentric vision, proprioception, and 3D scene geometry. We present sensVLA, a Vision-Language-Action (VLA) architecture that combines a Qwen3-2B Vision-Language Model (VLM) with a fully trainable transformer action expert trained by flow-matching velocity regression. sensVLA routes Bird's-Eye-View (BEV) features, extracted from fused front and rear lidar, directly to the action expert through a dedicated cross-attention pathway, while the VLM consumes front and rear RGB views to provide task-conditioned semantic context. This design decouples spatial grounding from linguistic reasoning while preserving interaction between both streams at decision time. The expert predicts six action dimensions: longitudinal velocity, steering, body-frame displacement, arm rate, and bucket rate. On a real-world dataset from a wheel loader, sensVLA reaches aggregate per-step parity with a strong camera-only baseline and reduces longitudinal velocity RMSE by 28% and displacement error by 9% on loading centric scenarios. It also degrades 29% less when the camera stream is corrupted or removed, evidencing that explicit spatial grounding improves accuracy and fault-tolerance for heavy equipment autonomy.
Vision-Language-Action (VLA) models commonly adapt pretrained Vision-Language Models (VLMs) to robot control by mapping visual observations and language instructions to continuous actions. Existing approaches typically take an action-insufficiency view, assuming that pretrained VLM latents either lack directly usable action information or should be shielded from action-learning signals. Against this view, our \textit{Quotient Theory for VLA} shows that pretrained VLM latents are not action-insufficient but action-sufficient: they already contain the information needed for control, yet remain overcomplete by distinguishing prompt-level variations that induce the same optimal action behavior. To operationalize this theory, we propose QuoVLA, a quotient-space framework for VLA that compresses pretrained VLM latents into action-sufficient representations. Specifically, QuoVLA instantiates this principle with a quantization module and a dual-branch design with relative temporal-complexity regularization, preserving action-relevant information while removing prompt-level redundancy. Extensive experiments across multiple benchmarks demonstrate that QuoVLA achieves strong performance, with particularly notable improvements in generalization under visual, linguistic, and environmental distribution shifts. Our code will be made publicly available.
Visual representations of VLA models remain unreliable for spatially precise robotic manipulation. We uncover that vision encoders in VLAs also exhibit attention artifacts previously documented in generic Vision Transformers, and further show that, in embodied policies, these artifacts are closely associated with spatial perception capabilities acquired during post-training. As the encoder learns task-relevant information such as object location, depth ordering, and local geometry, limited global-token capacity causes part of this information to spill into low-information patch tokens. We introduce AtVLA, a framework that inserts learnable register tokens into the visual encoder. Trained end-to-end using only embodied data and the original action objective, these registers emerge as dedicated carriers of embodied spatial information, while the remaining patch tokens recover clean and spatially faithful attention distributions crucial for precise target localization and fine-grained contact. Clean attention restores reliable localization, but cannot recover geometric details lost in low-resolution observations. AtVLA therefore couples attention rectification with uncertainty-gated local refinement. The action expert samples multiple action chunks and estimates uncertainty from their disagreement; only for uncertain predictions, action-conditioned attention rollout identifies the task-relevant region, which is cropped, re-encoded at high resolution, and appended to the cached prefix for refined action generation. Across LIBERO, SimplerEnv, and a challenging single-view real-world benchmark, AtVLA improves the average LIBERO success rate from 94.2% to 98.4% and real-world success from 46.5% to 69.0%. The cropping is triggered on approximately 30% of replanning steps, resulting in only 1.4-1.6x the total computation of the base model under the representative deployment setting.
Vision-language-action (VLA) models can describe scenes and reason about them in language, yet still struggle to ground their actions in the dense 3D world around them. Existing approaches either inject features from a frozen 3D foundation model without an objective that ensures the policy uses them, or constrain geometry with sparse box and map losses that provide no dense spatial signal. We introduce VLGA, the first vision-language-action model supervised to reconstruct the dense 3D world it drives through. VLGA introduces geometry as a fourth modality alongside vision, language, and action through a dedicated expert supervised by a per-pixel pointmap regression loss against LiDAR. Extensive experiments conducted on challenging nuScenes and Bench2Drive datasets for open-loop and closed-loop evaluations, respectively, show the superiority of VLGA over counterpart VLA methods. In particular, on open-loop nuScenes, VLGA sets a new state of the art among VLA methods without ego status, with the lowest L2 (0.50,m average) and 3-second collision rate (0.18%). On closed-loop Bench2Drive, VLGA attains the state-of-the-art driving score of 79.08, +0.71 over the strongest prior VLA, at comparable efficiency and comfort.