Back to Home

Agentic AI Driving Decision Engine

An autonomous vehicle decision-making engine powered by LangChain and Google's Gemini LLMs. This engine uses a tool-calling architecture to gather real-time simulated telemetry and perception data, reason about the driving environment, and output rigid, structured control commands.

Python 3.10LangChainGoogle GeminiPydanticpytest

Key Engineering Features

  • Reason (Thought): The agent receives a driving scenario (e.g., “Approaching an intersection at 15 m/s”) and determines what information is missing to make a safe decision.
  • Act (Tool Execution): The agent calls specific tools to interact with the simulated vehicle environment (telemetry_tool, perception_tool).
  • Observe (Tool Output): The agent receives the data from the tools and incorporates it into its reasoning context.
  • Final Decision: Once sufficient information is gathered, the agent processes the consolidated context through its system prompt and produces a strictly formatted JSON output adhering to a Pydantic schema.