dora-rs/dora

Dora is a Rust framework for building real-time robotics and AI applications. It is designed around the concept of dataflows: you describe your application as a directed graph of nodes, where each node does a specific job (reading a camera, running a detection model, sending motor commands), and Dora handles passing data between them efficiently. This graph is written in a YAML file, making it easy to visualize and reconfigure the pipeline without changing code. Performance is a central focus. The framework claims to be 10 to 17 times faster than the equivalent Python setup in ROS2, which is the standard middleware used in the robotics industry. It achieves this by using shared memory to pass large data between nodes on the same machine, skipping unnecessary copying, and using Apache Arrow as its internal data format, which avoids serialization overhead. Nodes running on different machines communicate automatically over a network protocol called Zenoh. Developing with Dora does not lock you into a single language. Nodes can be written in Rust, Python, C, or C++, and you can mix languages freely in one dataflow. Python nodes can be reloaded live while the application is running, which speeds up iteration. The framework also ships with a library of ready-made nodes for common robotics tasks: cameras, object detection with YOLO, large language models, and text-to-speech. The single command-line tool covers the full development cycle: running locally, deploying to a cluster of machines over SSH, inspecting live data streams, recording and replaying sessions for testing, and monitoring per-node CPU and memory usage across all machines in a terminal dashboard. It also includes a bridge for working with ROS2, the existing industry standard, so Dora nodes and ROS2 nodes can exchange messages. For production use, Dora supports per-node restart policies with backoff, dynamic addition and removal of nodes from a running system, optional soft real-time scheduling, and distributed tracing through OpenTelemetry. The framework is available via the Rust package registry and Python's pip, and installers are provided for macOS, Linux, and Windows.