DRIVESPATIAL: A Benchmark for Spatiotemporal Intelligence in VLMs for Autonomous Driving

• URL: http://arxiv.org/abs/2605.23176v1
• Date: Fri, 22 May 2026 02:52:06 GMT
• Title: DRIVESPATIAL: A Benchmark for Spatiotemporal Intelligence in VLMs for Autonomous Driving
• Abstract summary: We introduce DriveSpatial, a benchmark of 15.6K-verified QA pairs across 20 tasks from five large-scale autonomous driving datasets.<n>Unlike prior benchmarks DriveSpatial is generated from a dynamic multi-language scene graph that encodes object states, spatial relations, interactions, camera visibility, and temporal correspondences.<n>We find that the strongest model trails humans by 28.4 points, with Cognitive Scene Construction emerging as the key bottleneck.
• Score: 15.87281823665284
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Spatiotemporal intelligence in autonomous driving (AD) requires an agent to integrate multi-view observations into a coherent scene representation, maintain object continuity across viewpoints and time, and reason about spatial relations, interactions, and future dynamics. However, existing AD vision-language benchmarks largely focus on single-view, static, ego-centric, or single-source question answering, leaving it unclear whether current Vision-Language Models (VLMs) can truly construct and reason over dynamic driving scenes. We introduce DriveSpatial, a benchmark of 15.6K human-verified QA pairs across 20 tasks from five large-scale AD datasets. DriveSpatial evaluates four abilities: Cognitive Scene Construction, Multi-view Relational Understanding, Temporal Reasoning, and Generalization. Unlike prior benchmarks, DriveSpatial is generated from a dynamic multi-relational scene graph that encodes object states, spatial relations, interactions, camera visibility, and temporal correspondences, enabling QA pairs that enforce genuine cross-view and spatiotemporal reasoning. Evaluating 15 representative VLMs reveals a substantial human-model gap: the strongest model trails humans by 28.4 points, with Cognitive Scene Construction emerging as the key bottleneck. Further diagnostics show that language-only prompting is insufficient, while explicit BEV grounding consistently improves performance. These results suggest that current VLMs lack the scene-construction ability needed for reliable spatiotemporal driving intelligence. DriveSpatial and its construction pipeline will be released to support future research.

Related papers

• Towards Generalizable Visually Grounded Exploration of Household Devices [74.66117923766632]
  Key bottleneck of realizing general autonomous embodied agents lies in Generalizable Visually Grounded Exploration.<n>We introduce VGEBench, a benchmark designed to evaluate the generalizable visually grounded exploration capabilities of Vision-Language Models.<n>Unlike static datasets, we construct a Logic-Driven State Machine framework. This framework simulates multi-turn interaction loops, compelling agents to achieve goals by active visual perception and feedback-driven correction.
  arXiv  Detail & Related papers  (2026-09-01T07:42:45Z)
• MindVLA-U1: VLA Beats VA with Unified Streaming Architecture for Autonomous Driving [54.57163800903507]
  We present MindVLA-U1, the first unified streaming VLA architecture for autonomous driving.<n>A unified VLM backbone produces AR language tokens and flow-matching continuous action trajectories in a single forward pass over one shared representation.<n>On the long-tail WOD-E2E benchmark, MindVLA-U1 surpasses experienced human drivers for the first time.
  arXiv  Detail & Related papers  (2026-05-12T18:09:42Z)
• LatentPilot: Scene-Aware Vision-and-Language Navigation by Dreaming Ahead with Latent Visual Reasoning [51.969318585152116]
  LatentPilot exploits future observations during training as a valuable data source to learn action-conditioned visual dynamics.<n>We propose a flywheel-style training mechanism that iteratively collects on-policy trajectories and retrains the model to better match the agent's behavior distribution.<n> Experiments on R2R-CE, RxR-CE, and R2R-PE benchmarks achieve new SOTA results, and real-robot tests across diverse environments demonstrate LatentPilot's superior understanding of environment-action dynamics in scene.
  arXiv  Detail & Related papers  (2026-03-31T02:21:59Z)
• From Segments to Scenes: Temporal Understanding in Autonomous Driving via Vision-Language Model [13.666165551248701]
  Temporal understanding in autonomous driving (AD) remains a significant challenge.<n>Prior work has introduced datasets and benchmarks aimed at improving temporal reasoning.<n>No existing benchmark focuses exclusively on unique challenges of temporal understanding in ego-centric AD footage.
  arXiv  Detail & Related papers  (2025-12-04T21:57:10Z)
• OmniReason: A Temporal-Guided Vision-Language-Action Framework for Autonomous Driving [6.136909494522169]
  We propose OmniReason, which establishes temporal reasoning by jointly modeling dynamic 3D environments.<n>We also develop the OmniReason-Agent architecture, which integrates temporal memory for scene context.
  arXiv  Detail & Related papers  (2025-08-31T10:34:44Z)
• STRIDE-QA: Visual Question Answering Dataset for Spatiotemporal Reasoning in Urban Driving Scenes [5.685235562999083]
  STRIDE-QA is the largest visual question answering dataset fortemporal reasoning in urban driving.<n>It supports both object-centric and ego-centric reasoning through spatial localization and temporal prediction.<n>Our benchmarks demonstrate that existing Vision-Language Models (VLMs) struggle to achieve near-zero scores on prediction consistency.
  arXiv  Detail & Related papers  (2025-08-14T07:57:06Z)
• VLM4D: Towards Spatiotemporal Awareness in Vision Language Models [66.833085504228]
  We introduce V4DLM, the first benchmark specifically designed to evaluate visual language models (VLMs)<n>Our benchmark comprises diverse real-world and synthetic videos accompanied by carefully curated question-answer pairs.<n>We identify significant performance gaps compared to human baselines, highlighting fundamental deficiencies in existing models.
  arXiv  Detail & Related papers  (2025-08-04T06:06:06Z)
• STSBench: A Spatio-temporal Scenario Benchmark for Multi-modal Large Language Models in Autonomous Driving [16.602141801221364]
  STSBench is a framework to benchmark holistic understanding of vision-language models (VLMs) for autonomous driving.<n>The benchmark features 43 diverse scenarios spanning multiple views, resulting in 971 human-verified multiple-choice questions.<n>A thorough evaluation uncovers shortcomings in existing models' ability to reason about fundamental traffic dynamics in complex environments.
  arXiv  Detail & Related papers  (2025-06-06T16:25:22Z)
• Natural Reflection Backdoor Attack on Vision Language Model for Autonomous Driving [55.96227460521096]
  Vision-Language Models (VLMs) have been integrated into autonomous driving systems to enhance reasoning capabilities.<n>We propose a natural reflection-based backdoor attack targeting VLM systems in autonomous driving scenarios.<n>Our findings uncover a new class of attacks that exploit the stringent real-time requirements of autonomous driving.
  arXiv  Detail & Related papers  (2025-05-09T20:28:17Z)
• NuScenes-SpatialQA: A Spatial Understanding and Reasoning Benchmark for Vision-Language Models in Autonomous Driving [10.41584658117874]
  We propose NuScenes-SpatialQA, the first large-scale ground-truth-based Question-Answer (QA) benchmark designed to evaluate the spatial understanding and reasoning capabilities of Vision-Language Models (VLMs) in autonomous driving.<n>Built upon the NuScenes dataset, the benchmark is constructed through an automated 3D scene graph generation pipeline and a QA generation pipeline.<n>Using this benchmark, we conduct extensive experiments on diverse VLMs, including both general and spatial-enhanced models, providing the first comprehensive evaluation of their spatial capabilities in autonomous driving.
  arXiv  Detail & Related papers  (2025-04-04T04:43:10Z)
• A Spatio-Temporal Multilayer Perceptron for Gesture Recognition [70.34489104710366]
  We propose a multilayer state-weighted perceptron for gesture recognition in the context of autonomous vehicles. An evaluation of TCG and Drive&Act datasets is provided to showcase the promising performance of our approach. We deploy our model to our autonomous vehicle to show its real-time capability and stable execution.
  arXiv  Detail & Related papers  (2022-04-25T08:42:47Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.