Dilica
Mosaico is an open-source data platform that helps robotics teams turn petabyte-scale sensor data into training-ready assets. It runs entirely on-premise, ingesting formats like ROS bags and MCAP through a single interface, and enables filtering by physical sensor values across entire datasets. The platform provides a Python SDK with push, find, and stream primitives to cover the full data lifecycle without SQL or boilerplate.
- Artificial Intelligence
- Developer Tools
- Robotics
- Software Only
Funding
Founders
Product
Problem
Robotics teams generate petabytes of heterogeneous sensor data from custom rigs, ROS-based systems, and edge devices, but lack purpose-built infrastructure to manage it. Existing workarounds are fragile, require extensive format conversion, and make it difficult to locate specific events or time windows across massive datasets, slowing down the path from raw data to training-ready assets.
Solution
Mosaico provides an open-source, on-premise data platform designed specifically for robotics and physical AI workloads. It handles the entire data lifecycle—ingestion, ontology, and query—entirely on the user's own infrastructure, ensuring data never leaves their servers or edge hardware. The platform ingests diverse formats like ROS bags, MCAP, and custom sensor streams through a single structured interface with automatic schema translation. A Python SDK offers three core primitives—push, find, and stream—enabling teams to write code for new ideas without dealing with SQL, schemas, or boilerplate. Users can filter by physical sensor values across entire datasets, such as acceleration spikes or GPS drops, and retrieve exact timestamp windows instead of scrubbing through files manually.
Target Audience
Primary users are robotics engineering teams, physical AI researchers, and data infrastructure engineers at companies building autonomous systems, who need to manage and query large-scale sensor data on their own infrastructure.
Features
- On-premise deployment across all layers (ontology, ingestion, query) on user servers, cloud accounts, or edge hardware
- Sensor-value-based filtering across entire datasets to locate events like acceleration spikes, GPS drops, or joint overloads
- Automatic schema translation for ROS bags, MCAP, and custom sensor formats through a single structured interface
- Python SDK with push, find, and stream primitives covering the full data lifecycle without SQL or boilerplate
- Session-level error policies (e.g., delete on error) for robust ingestion workflows
- Metadata tagging for sequences, enabling source tracking (e.g., "sensor_rig_v3") and organized data management