DriveMCP: An MCP-Powered Agentic Driver Assistance System
Submitted to IEEE Transactions on Intelligent Vehicles, February 2026
Grounded in a CARLA camera and LiDAR perception stack (RGB detection, range refinement, Kalman tracking) feeding a typed world state with per-field confidence. Evaluated against VLM-direct, RAG and no-arbiter baselines under injected perception and CAN faults. The telemetry server in this stack is open source: MCP-CAN.
Recommended citation: Nadiri, F., & Rad, A. B. (2026, under review). "DriveMCP: An MCP-Powered Agentic Driver Assistance System." IEEE Transactions on Intelligent Vehicles.