Research map / Vision-language and multimodal / Visual question answering
DriveLM: Driving with Graph Visual Question Answering
Chonghao Sima, Katrin Renz, Kashyap Chitta, Li Chen, Zhang Hanxue, Chengen Xie, Jens Beißwenger, Ping Luo, Andreas Geiger, Hongyang Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add graph context (all P1-3 QAs) to behavior prediction (ablation) | DriveLM-Agent · Waymo zero-shot | behavior speed accuracy: 43.90 -> 54.29 | Improved |
| use chain context (only P3) instead of full graph (ablation) | DriveLM-Agent · Waymo zero-shot | behavior speed accuracy: 43.90 -> 41.28 | Hurt |
| add graph or chain context for behavior (ablation) | DriveLM-Agent · DriveLM-nuScenes (in-domain) | ADE: 1.39 (None) vs 1.74 (Graph) | No clear effect |
| predict behavior as intermediate step before motion | BLIP-RT-2 / DriveLM-Agent · DriveLM-nuScenes | ADE: 2.63 -> 1.39 | Improved |
| add prediction QA (P2-1) context to perception-only context (ablation) | DriveLM-Agent behavior · DriveLM-nuScenes | behavior accuracy: 53.94 -> 58.82 | Improved |
| add planning QAs on top of prediction QAs (ablation) | DriveLM-Agent behavior · DriveLM-nuScenes | behavior accuracy: 57.70 -> 58.95 (P3-1), 57.49 with all | No clear effect |
| add graph context to P1-3 QA (ablation) | DriveLM-Agent · DriveLM-nuScenes P1-3 | SPICE: 42.56 -> 49.54 | Improved |
| add graph context to zero-shot BLIP-2 (ablation) | BLIP-2 · DriveLM-CARLA P1-3 | GPT score: 46.37 -> 55.03 | Improved |
| add graph context to zero-shot BLIP-2 (ablation) | BLIP-2 · DriveLM-CARLA P1-3 | SPICE: 10.46 -> 10.30 | No clear effect |
| fine-tune on DriveLM data vs zero-shot BLIP-2 | BLIP-2 · DriveLM-nuScenes P1-3 | GPT score: 42.97 -> 71.39 | Improved |
| use full graph context in Waymo zero-shot planning (ablation) | DriveLM-Agent · Waymo zero-shot | ADE: 2.76 -> 2.63 | Improved |
Closest papers by meaning
- DriveSpatial: A Benchmark for Spatiotemporal Intelligence in VLMs for Autonomous Driving (NeurIPS 2026 · similarity 0.91)
- NuScenes-QA: A Multi-Modal Visual Question Answering Benchmark for Autonomous Driving Scenario (AAAI 2024 · similarity 0.91)
- More Than Meets the Eye? Uncovering the Reasoning-Planning Disconnect in Training Vision-Language Driving Models (NeurIPS 2026 · similarity 0.90)
- STSBench: A Spatio-temporal Scenario Benchmark for Multi-modal Large Language Models in Autonomous Driving (NeurIPS 2025 · similarity 0.89)
- DriveAgent-R1: Advancing VLM-based Autonomous Driving with Active Perception and Hybrid Thinking (ICLR 2026 · similarity 0.89)
- nuReasoning: A Reasoning-Centric Dataset and Benchmark for Long-Tail Autonomous Driving (NeurIPS 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
