Pixel Robotics turns rosbags into queryable SQL databases for LLM analysis
No custom tools needed—LLMs query robot logs directly via SQL joins.
Pixel Robotics, which builds AMRs for pallet transport, has detailed a fresh technique for using LLMs to analyze ROS 2 rosbag data. Their approach leans on mcap-mcp-server, an MCP server that stores rosbag contents in a database. This design lets large language models write and execute complex SQL queries, including joins across different message types, for fast, flexible analysis of recorded robot sessions. Because the LLM already knows SQL, there's no need to build special-purpose analysis tools for each kind of data.
Beyond that, the team extended their Nav2 behavior-tree logging so the SQL database captures behavior-tree state transitions, making it possible to query what the robot was doing and why at any timestamp. They also published an agent skill that automates the environment setup, so engineers can spin up the whole pipeline quickly. The result is an efficient workflow: ask a question in natural language, get an answer from structured queries across trajectory, laser scan, tf, and behavior-tree logs.
- mcap-mcp-server stores rosbags in a database, enabling complex SQL joins for fast queries
- Nav2 behavior-tree logging was extended to make robot decisions queryable alongside sensor data
- An agent skill is shared to automate setup, letting teams start with LLM-driven rosbag analysis immediately
Why It Matters
This gives robotics engineers a scalable, LLM-friendly way to extract insights from complex log data without bespoke tooling.