DriveFusion/DriveFusionQA

VISIONPricing:Input $0.32 / Cached $0.016 / Output $1.6Concurrent Unit Cost:1Model Size:3BQuant:BF16Context Size:32kPublished:Jul 1, 2025License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

DriveFusionQA is a 3 billion parameter Vision-Language Model (VLM) developed by DriveFusion, built upon the Qwen2.5-VL architecture. It is specifically fine-tuned for autonomous driving scenarios, excelling at interpreting complex driving scenes and explaining vehicle decision-making. This model bridges the gap between raw sensor data and human-understandable reasoning, making it ideal for safety analysis and simulation in autonomous systems.

Loading preview...

DriveFusionQA: Autonomous Driving VLM

DriveFusionQA is a specialized Vision-Language Model (VLM) from DriveFusion, fine-tuned on the Qwen2.5-VL architecture to understand and explain complex autonomous driving scenarios. Unlike general-purpose VLMs, DriveFusionQA focuses on providing human-understandable reasoning for vehicle decisions, making it a critical tool for safety analysis and interactive driving support.

Key Capabilities

  • Scenario Explanation: Identifies traffic participants, road signs, and environmental hazards within driving scenes.
  • Decision Reasoning: Justifies driving actions, such as "Braking due to a pedestrian entering the crosswalk."
  • Multi-Dataset Expertise: Trained on a unified pipeline of world-class driving benchmarks including LingoQA, DriveGPT4 + BDD-X, and DriveLM.
  • Interactive Dialogue: Supports multi-turn conversations about road safety and navigation.

Performance

DriveFusionQA demonstrates significant improvements over its base model, Qwen2.5-VL, particularly in driving-related language metrics. It achieves a Lingo-Judge score of 53.2, reflecting superior ability to generate human-aligned driving reasoning, compared to the base model's 38.1.

Intended Use Cases

  • Safety Analysis: Generating natural language reports for dashcam footage and near-miss events.
  • Training & Simulation: Providing ground-truth explanations for AI driver training.
  • Interactive Assistants: Assisting human operators or passengers with scene descriptions.

Limitations

  • May occasionally misinterpret distant objects or complex social traffic cues (hallucination).
  • Performance can vary in regions or weather conditions not well-represented in training data.
  • Designed for reasoning and explanation, not for direct vehicle control.

Finetunes
1 models