Research map / Vision-language and multimodal / Robot action and navigation
DrivePI: Spatial-aware 4D MLLM for Unified Autonomous Driving Understanding, Perception, Prediction and Planning
Zhe Liu, Runhui Huang, Rui Yang, Siming Yan, Zining Wang, Lu Hou, Di Lin, Xiang Bai, Hengshuang Zhao
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 the vision heads (occupancy, flow, planning) to the text head (ablation) | DrivePI-0.5B · nuScenes-QA val | accuracy: 61.2 -> 60.7 | Hurt |
| add the text head to the vision-only model (ablation) | DrivePI-0.5B · OpenOcc val | RayIoU: 47.5 -> 49.3 | Improved |
| add the text head to the vision-only model (ablation) | DrivePI-0.5B · OpenOcc val | mAVE: 0.69 -> 0.51 | Improved |
| add the text head to the vision-only model (ablation) | DrivePI-0.5B · nuScenes planning | L2 (m): 1.02 -> 0.50 | Improved |
| add the text head to the vision-only model (ablation) | DrivePI-0.5B · nuScenes planning | collision rate: 0.39 -> 0.38 | No clear effect |
| raise occupancy and flow loss weights from 0.2 to 1.0 (ablation) | DrivePI-0.5B · OpenOcc val | RayIoU: 48.1 -> 49.3 | Improved |
| raise occupancy and flow loss weights from 0.2 to 1.0 (ablation) | DrivePI-0.5B · OpenOcc val | mAVE: 0.57 -> 0.51 | Improved |
| raise occupancy and flow loss weights from 0.2 to 1.0 (ablation) | DrivePI-0.5B · nuScenes planning | collision rate: 0.19 -> 0.38 | Hurt |
| scale occupancy QA data from 28k to 560k (ablation) | DrivePI 3B text head · Occupancy class QA | accuracy: 14.3 -> 59.2 | Improved |
| add occupancy-flow, action and nuScenes-QA data to 420k occupancy QA (ablation) | DrivePI 3B text head · Occupancy class QA | accuracy: 59.2 -> 59.8 | Improved |
| scale model from 0.5B to 3B (ablation) | DrivePI · Occupancy status QA | accuracy: 86.0 -> 88.6 | Improved |
| scale model from 0.5B to 3B (ablation) | DrivePI · Planning | L2 (m): 0.79 -> 0.86 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- OpenDriveVLA-7B — OpenDriveVLA: Towards End-to-end Autonomous Driving with Large Vision Language Action Model (AAAI 2026) · better on 4 of 4 shared results
- ORION — ORION: A Holistic End-to-End Autonomous Driving Framework by Vision-Language Instructed Action Generation (ICCV 2025) · better on 1 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SpaceDrive: Infusing Spatial Awareness into VLM-based Autonomous Driving (CVPR 2026 · similarity 0.90)
- VGGDrive: Empowering Vision-Language Models with Cross-View Geometric Grounding for Autonomous Driving (CVPR 2026 · similarity 0.90)
- S4-Driver: Scalable Self-Supervised Driving Multimodal Large Language Model with Spatio-Temporal Visual Representation (CVPR 2025 · similarity 0.90)
- Explicit Geometric Chain-of-Thought for Vision-Language-Action in Autonomous Driving (NeurIPS 2026 · similarity 0.89)
- MAPLM: A Real-World Large-Scale Vision-Language Benchmark for Map and Traffic Scene Understanding (CVPR 2024 · similarity 0.89)
- FutureSightDrive: Thinking Visually with Spatio-Temporal CoT for Autonomous Driving (NeurIPS 2025 · 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.
