⬅️ Back
Interactive Mode

Driver Behavior Modeling and Use of Automation

Tags

Scope:

Behavior Modeling

Keywords:

Context-aware ModelingExplainable AILane Change PredictionForward Collision WarningMulti-modal FusionAttention MechanismsNaturalistic Driving Study
📋

Project Overview

Comprehensive research program at MIT AgeLab developing context-aware driver behavior models using explainable AI methods. The core approach focuses on understanding triads of vehicle-driver-environment interactions in partial automation scenarios.

  • Context-aware modeling framework integrating vehicle dynamics, driver state, and environmental factors
  • Multi-modal naturalistic driving data analysis using cameras, CAN bus, GPS, and IMU sensors
  • Explainable AI methods for understanding human-automation interaction patterns
  • Multiple research directions covering lane changes, collision warnings, transfer of control, and pedestrian interactions
🔬

Methodology

Employs naturalistic driving data from MIT-AVT dataset combined with advanced machine learning methods to understand context-dependent driver behavior in automated driving scenarios.

  • Naturalistic driving data collection: multi-perspective cameras (in-cabin face/body, forward view), CAN bus, GPS, IMU
  • Context-aware modeling: triads of vehicle-driver-environment interactions with XAI explainability
  • Multiple AI approaches: CNN-RNN with attention mechanisms, Graph Neural Networks (GNN), Large Language Models (LLM)
  • Cross-scenario analysis: highway vs local roads, different automation levels, various traffic conditions
Flowchart for lane change prediction

Flowchart for lane change prediction

Results & Impact

Published and ongoing work across naturalistic driving, interviews, multimodal modeling, and explainable AI provides a connected view of human-automation interaction.

  • Lane Change Prediction: 87% F1-score using multi-modal spatio-temporal attention networks
  • Forward Collision Warning Analysis: Context-dependent effectiveness across road types and traffic scenarios
  • Transfer of Control: Published a 2026 context-aware GNN and explainable AI framework for prediction and design
  • Driver Interviews: Published a 2026 LLM-agent workflow for comparing driver experience across partial automation systems
  • Current Directions: Multimodal driver-pedestrian negotiation, FCW response modeling, and anomaly detection for transfers of control
🔗

Related Resources

External Links:

🔗 MIT Advanced Vehicle Technology Consortium

Publications:

End-to-End Spatio-Temporal Attention-Based Lane-Change Intention Prediction from Multi-Perspective Cameras

Driver Behavior in Response to Forward Collision Warnings Considering Driving Context

From Prediction to Design: Using Context-Aware Graph Neural Networks and Explainable AI to Anticipate Transfer-of-Control

Large-Language-Model-Agent-Enabled Analysis of Semi-Structured Interviews: Comparing Driver Experience Across Partial Automation Systems