# Trajectory Prediction on nuScenes (L2 Error and Collision Rate) Source: [Trajectory Prediction on nuScenes (L2 Error and Collision Rate)]() Collection index: [llms.txt]() Data updated: 2026-07-03T06:59:36.580Z Evaluation of trajectory prediction models on the nuScenes dataset, specifically reporting L2 displacement errors (in meters) and collision rates across 1, 2, and 3-second prediction horizons. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Trajectory prediction | | Dataset (including any stated split) | nuScenes | | Metric | L2 Error (m) 1s | | Unit | SCALAR | | Direction | Lower is better | | Results loaded | 27 of 27 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [L2 Error (m) 1s](): Lower is better; unit: SCALAR. - [L2 Error (m) 2s](): Lower is better; unit: SCALAR. - [L2 Error (m) 3s](): Lower is better; unit: SCALAR. - [L2 Error (m) Avg.](): Lower is better; unit: SCALAR. - [Collision Rate 1s](): Lower is better; unit: PERCENT. - [Collision Rate 2s](): Lower is better; unit: PERCENT. - [Collision Rate 3s](): Lower is better; unit: PERCENT. - [Collision Rate Avg.](): Lower is better; unit: PERCENT. ## Recorded results | Method | L2 Error (m) 1s | Result date | Report paper | | --- | --- | --- | --- | | DriveTeach-VLA-3B | 0.12 | Not recorded | 6a474209143ebdcacacc5c1f | | Impromptu-VLA-3B | 0.13 | Not recorded | 6a474209143ebdcacacc5c1f | | Impromptu-VLA-7B | 0.13 | Not recorded | 6a474209143ebdcacacc5c1f | | OmniDrive | 0.14 | Not recorded | 6a474209143ebdcacacc5c1f | | EMMA | 0.14 | Not recorded | 6a474209143ebdcacacc5c1f | | Ego-MLP\* | 0.15 | Not recorded | 6a474209143ebdcacacc5c1f | | DriveVLM-Dual | 0.15 | Not recorded | 6a474209143ebdcacacc5c1f | | EMMA (random init) | 0.15 | Not recorded | 6a474209143ebdcacacc5c1f | | BEV-Planner | 0.16 | Not recorded | 6a474209143ebdcacacc5c1f | | VAD | 0.17 | Not recorded | 6a474209143ebdcacacc5c1f | | DriveVLM | 0.18 | Not recorded | 6a474209143ebdcacacc5c1f | | PRIX | 0.26 | Not recorded | 68fa7a2a1e209131cea81758 | | GPT-4o | 0.28 | Not recorded | 6a474209143ebdcacacc5c1f | | Claude-3.7-Sonnet | 0.28 | Not recorded | 6a474209143ebdcacacc5c1f | | SparseDrive | 0.29 | Not recorded | 68fa7a2a1e209131cea81758 | | DiffusionDrive | 0.31 | Not recorded | 68fa7a2a1e209131cea81758 | | Gemini-2.5-Pro | 0.37 | Not recorded | 6a474209143ebdcacacc5c1f | | VAD | 0.41 | Not recorded | 68fa7a2a1e209131cea81758 | | UniAD | 0.42 | Not recorded | 6a474209143ebdcacacc5c1f | | UniAD | 0.45 | Not recorded | 68fa7a2a1e209131cea81758 | | Qwen-2.5-VL-7B-Instruct | 0.46 | Not recorded | 6a474209143ebdcacacc5c1f | | LLaMA-3.2-11B-Vision-Instruct | 0.52 | Not recorded | 6a474209143ebdcacacc5c1f | | DeepSeek-VL2-16B | 0.66 | Not recorded | 6a474209143ebdcacacc5c1f | | OccNet | 1.29 | Not recorded | 68fa7a2a1e209131cea81758 | | ST-P3 | 1.33 | Not recorded | 68fa7a2a1e209131cea81758 | | Qwen2-VL-7B-Instruct | 1.45 | Not recorded | 6a474209143ebdcacacc5c1f | | LLaVA-1.6-Mistral-7B | 1.49 | Not recorded | 6a474209143ebdcacacc5c1f | ### DriveTeach-VLA-3B L2 Error (m) 1s: 0.12. Result ID: 6a475db960b88bb33992566d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Specialized Driving Models","Model Size":"3B"}. This record does not assert independent reproduction. ### Impromptu-VLA-3B L2 Error (m) 1s: 0.13. Result ID: 6a475db960b88bb33992566b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Specialized Driving Models","Model Size":"3B"}. This record does not assert independent reproduction. ### Impromptu-VLA-7B L2 Error (m) 1s: 0.13. Result ID: 6a475db960b88bb33992566c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Specialized Driving Models","Model Size":"7B"}. This record does not assert independent reproduction. ### OmniDrive L2 Error (m) 1s: 0.14. Result ID: 6a475db960b88bb339925667. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Specialized Driving Models"}. This record does not assert independent reproduction. ### EMMA L2 Error (m) 1s: 0.14. Result ID: 6a475db960b88bb33992566a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Specialized Driving Models"}. This record does not assert independent reproduction. ### Ego-MLP\* L2 Error (m) 1s: 0.15. Result ID: 6a475db960b88bb339925665. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Category":"Training-based Driving Specialists (Existing Methods)"}. This record does not assert independent reproduction. ### DriveVLM-Dual L2 Error (m) 1s: 0.15. Result ID: 6a475db960b88bb339925668. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Specialized Driving Models"}. This record does not assert independent reproduction. ### EMMA (random init) L2 Error (m) 1s: 0.15. Result ID: 6a475db960b88bb339925669. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Specialized Driving Models","initialization":"random init"}. This record does not assert independent reproduction. ### BEV-Planner L2 Error (m) 1s: 0.16. Result ID: 6a475db960b88bb339925664. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Category":"Training-based Driving Specialists (Existing Methods)"}. This record does not assert independent reproduction. ### VAD L2 Error (m) 1s: 0.17. Result ID: 6a475db960b88bb339925663. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Category":"Training-based Driving Specialists (Existing Methods)"}. This record does not assert independent reproduction. ### DriveVLM L2 Error (m) 1s: 0.18. Result ID: 6a475db960b88bb339925666. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Specialized Driving Models"}. This record does not assert independent reproduction. ### PRIX L2 Error (m) 1s: 0.26. Result ID: 69ddf80f47f34ce924c5508a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Input":"Camera","Backbone":"ResNet-50","FPS":"11.2"}. This record does not assert independent reproduction. ### GPT-4o L2 Error (m) 1s: 0.28. Result ID: 6a475db960b88bb33992565a. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Generalist VLMs"}. This record does not assert independent reproduction. ### Claude-3.7-Sonnet L2 Error (m) 1s: 0.28. Result ID: 6a475db960b88bb33992565b. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Generalist VLMs"}. This record does not assert independent reproduction. ### SparseDrive L2 Error (m) 1s: 0.29. Result ID: 69ddf80f47f34ce924c55088. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Input":"Camera","Backbone":"ResNet-50","FPS":"9.0"}. This record does not assert independent reproduction. ### DiffusionDrive L2 Error (m) 1s: 0.31. Result ID: 69ddf80f47f34ce924c55089. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Input":"Camera","Backbone":"ResNet-50","FPS":"8.2"}. This record does not assert independent reproduction. ### Gemini-2.5-Pro L2 Error (m) 1s: 0.37. Result ID: 6a475db960b88bb33992565c. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Generalist VLMs"}. This record does not assert independent reproduction. ### VAD L2 Error (m) 1s: 0.41. Result ID: 69ddf80f47f34ce924c55087. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Input":"Camera","Backbone":"ResNet-50","FPS":"4.5"}. This record does not assert independent reproduction. ### UniAD L2 Error (m) 1s: 0.42. Result ID: 6a475db960b88bb339925662. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Model Category":"Training-based Driving Specialists (Existing Methods)"}. This record does not assert independent reproduction. ### UniAD L2 Error (m) 1s: 0.45. Result ID: 69ddf80f47f34ce924c55085. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Input":"Camera","Backbone":"ResNet-101","FPS":"1.8"}. This record does not assert independent reproduction. ### Qwen-2.5-VL-7B-Instruct L2 Error (m) 1s: 0.46. Result ID: 6a475db960b88bb339925661. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Generalist VLMs","Model Size":"7B"}. This record does not assert independent reproduction. ### LLaMA-3.2-11B-Vision-Instruct L2 Error (m) 1s: 0.52. Result ID: 6a475db960b88bb339925660. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Generalist VLMs","Model Size":"11B"}. This record does not assert independent reproduction. ### DeepSeek-VL2-16B L2 Error (m) 1s: 0.66. Result ID: 6a475db960b88bb33992565f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Generalist VLMs","Model Size":"16B"}. This record does not assert independent reproduction. ### OccNet L2 Error (m) 1s: 1.29. Result ID: 69ddf80f47f34ce924c55086. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Input":"Camera","Backbone":"ResNet-50","FPS":"2.6"}. This record does not assert independent reproduction. ### ST-P3 L2 Error (m) 1s: 1.33. Result ID: 69ddf80f47f34ce924c55084. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Input":"Camera","Backbone":"EffNet-b4","FPS":"1.6"}. This record does not assert independent reproduction. ### Qwen2-VL-7B-Instruct L2 Error (m) 1s: 1.45. Result ID: 6a475db960b88bb33992565e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Generalist VLMs","Model Size":"7B"}. This record does not assert independent reproduction. ### LLaVA-1.6-Mistral-7B L2 Error (m) 1s: 1.49. Result ID: 6a475db960b88bb33992565d. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Model Category":"Generalist VLMs","Model Size":"7B"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.