WorkStackAboutContact
← All Projects
Independent ProjectIn ProgressAutonomous Vehicles · Robotics

AutonomIQ —
Autonomous
Safety Intelligence

Multi-agent AI platform combining reinforcement learning, counterfactual reasoning, and LangGraph orchestration to generate stress-test scenarios from real AV incidents, train and evaluate autonomous policies, and produce structured safety audit reports for vehicles and robots.

↗ GitHub

Overview

AutonomIQ converts real autonomous vehicle incident data into stress-test scenarios and produces safety audit reports. The system ingests real incidents, generates 400K+ scenario variants through systematic parameter expansion, trains three RL policies (rule-based, DQN, PPO), evaluates them against all scenarios, and explains failures through counterfactual simulation. A six-agent LangGraph pipeline orchestrates the full flow from raw incident to structured safety passport PDF.

Architecture

Incident Intelligence
Claude API parses raw incidents (structured or free text) and extracts parameters. Feeds into scenario expansion.
Scenario Synthesis
One incident becomes 100 variants through parametric expansion (weather, speed, time of day, actors). 4,000 real incidents scale to 400,000 test scenarios.
Policy Training
Three RL agents trained in highway-env: rule-based baseline, DQN, and PPO (primary). Each tested against full scenario bank.
Counterfactual Analysis
For every failure, replay the scenario with masked policy action. Substitute alternatives and measure collision probability improvement.

Tech Stack

LangGraphClaude APIPPODQN Stable Baselines3highway-envXGBoostFastAPI ReportLabReact 18DockerHuggingFace Spaces
Back to
All Projects
→