Sensor Log Analysis for Autonomous Vehicles

• 

  Sensor Graph Genrated

• 

  UI

Inspiration

Autonomous systems rely on hundreds of sensors, ensuring safety and efficiency. However, when failures occur, identifying the root cause is time consuming and complex, often requiring manual log analysis. I wanted to build a solution that automates failure detection, analyzes dependencies, and enables real-time insights using graph analytics and natural language querying using gen AI.

What it does

1. Detects sensor failures in real-time from autonomous vehicle logs.
2. Analyzes dependencies between sensors using graph analytics.
3. Enables natural language querying for insights using LLMs (Gemini & Groq).
4. Visualizes sensor networks, highlighting failing vs. operational sensors.
5. Routes queries to cuGraph, NetworkX, or ArangoDB for the best results.

How we built it

1. Data Ingestion: Processed raw CSV logs to extract sensor readings.
2. Graph Construction: Converted sensor relationships into a graph structure.
3. Database Integration: Stored graph data in ArangoDB for efficient querying.
4. Querying: Used LLMs to process natural language queries to AQL, Networkx queries and cuGraph. Used LLM to select best answer out of three.
5. Graph Analytics: Applied cuGraph & NetworkX for failure detection & shortest path analysis. Interactive UI: Built a Streamlit UI for uploading logs, visualizing failures, and querying insights.

Challenges we ran into

1. Handling large sensor datasets → Optimized queries & used caching for performance.
2. Ensuring LLM query accuracy → Built a query routing system to select cuGraph, NetworkX, or AQL (Best answer from AQL, cuGraph, NetworkX) dynamically also used LLM to select best answer out of three.
3. LLM API limit →We used Gemni to create Queries for networkX and AQL. We used Grok for deciding best response for the output answer.
4. Real time failure detection → Managed sensor timestamps to track failures effectively.

Accomplishments that we're proud of

1. Developed a real time failure detection system using graphs.
2. Integrated multi agent AI capabilities for natural language query processing.
3. Optimized graph analytics with cuGraph, enabling fast dependency analysis.
4. Built an intuitive UI, making failure insights accessible to non-technical users.

What we learned

1. Graph analytics is powerful for failure detection – uncovering sensor dependencies beyond simple threshold checks.
2. LLMs can enhance real-time analytics – when properly structured with query routing.
   1. ArangoDB is a great choice for scalable graph-based storage & querying.
   2. Performance optimization is key – caching, indexing, and query selection improve real time insights.

What's next for Sensor Log Analysis for Autonomous Vehicles

1. Enhanced predictive maintenance using ML models for failure forecasting.
2. Multi modal AI integration – adding voice & image based diagnostics for richer insights.
3. Deploying on edge devices for real-time in-vehicle diagnostics & alerts.

Built With

• arangodb
• cugraph
• langchain
• networkx
• python
• streamlit