Author – Aryan Pandey, Priyadarshi Uttpal and Sanket Poojary

NeuroFCW is an advanced Forward Collision Warning (FCW) system powered by Generative AI, Neural Networks, and a Graph-RAG architecture. It automates code generation, test case creation, and continuous validation for FCW systems, ensuring robustness, precision, and real-world readiness.

Below is the high-level system architecture overview for NeuroFCW:

1. Document Segmentation and Preprocessing

   • Breaks down input documents into manageable segments for efficient processing.

2. Graph-RAG Knowledge Retrieval

   • Stores and retrieves safety standards, MISRA guidelines, test cases, and code examples using Neo4j Graph Knowledge Base.

3. Graph-RAG Code Generation

   • Generates FCW code compliant with MISRA standards using:
     • Contextual data retrieval
     • Code generation with a Large Language Model (LLM) API (Llama-3.3-70b pre-trained LLM)

4. Graph-RAG Test Case Generation

   • Automatically generates comprehensive test cases with relevant test patterns.

5. Fine-Tuned YOLO Model

   • Handles object detection and computes parameters for FCW validation.

6. Validation & Continuous Improvement

   • Failed test cases are logged and used to improve future FCW code.

7. Performance Logging

   • Logs critical metrics such as detection accuracy, processing speed, and code validation success rates to ensure real-world deployment readiness.

• AI/ML: YOLOv11, LangChain, OpenAI APIs
• Databases: Neo4j, SQLite
• Programming Languages: Python
• DevOps Tools: Jenkins, GitHub Actions, CARLA Simulation
• Frameworks: Large Language Models (LLMs), Graph-RAG

The system operates in the following steps:

1. Input Documents → Segmented into manageable chunks.
2. Knowledge Base Processing → Neo4j Aura graph database retrieves relevant guidelines and safety standards.
3. Code Generation → LLM APIs generate FCW code tailored to input requirements with MISRA compliance.
4. Test Case Generation → Retrieves patterns and validates with YOLO.
5. Validation → Test results logged, and the FCW system is updated continuously
6. Perfromance metrics → Critical KPIs (accuracy, processing speed, and anomaly handling) are logged for analysis.

• MISRA-Compliant FCW Code
• Comprehensive Test Cases
• Validated FCW Code Packages for Deployment

• Multi-Sensor Fusion with LiDAR, Radar, and Camera Systems
• Enhanced Fine-Tuning of LLMs for Anomaly Handling
• Reinforcement Learning for ADAS Decision-Making
• Energy-Efficient ML Models using lightweight and quantized ML models

• Aryan Pandey
• Priyadarshi Uttpal
• Sanket Poojary

1. Clone the repository:

   git clone https://github.com/ltd-ARYAN-pvt/NeuroFCW.git

2. Install required dependencies:

   pip install -r requirements.txt

3. Run the main program:

   python app.py

For more information, reach me at:

• 📧 aryan2002pandeythrgrt@gmail.com
• 📧 sanketpoojary003@gmail.com
• 📧 priyadarshiutpal06@gmail.com