UAVCAN


UAVCAN is a lightweight protocol designed for reliable intra-vehicle communications using various communications transports, originally destined for CAN bus but targeting various network types in subsequent revisions.

History

The first RFC broadly outlining the general ideas that would later form the core design principles of UAVCAN was published in early 2014. It was a response to the perceived lack of adequate technology that could facilitate robust real-time intra-vehicular data exchange between distributed components of modern intelligent vehicles.
Since the original RFC, the protocol has been through three major design iterations, which culminated in the release of the first long-term stable revision in 2020 labelled UAVCAN v1.0. In the meantime, the protocol has been deployed in numerous diverse systems including unmanned aerial vehicles, spacecraft, underwater robots, racing cars, general robotic systems, and micro mobility vehicles.
UAVCAN v1.0 is positioned by its developers as a highly deterministic, safety-oriented alternative to high-level publish-subscribe frameworks such as DDS or the computation graph of ROS, which is sufficiently compact and simple to be usable in deeply embedded high-integrity applications. UAVCAN has been shown to be usable with bare metal microcontrollers equipped with as little as 32K ROM and 8K RAM.
The protocol is open and can be reused freely without approval or licensing fees. The development of the core standard and its reference implementations is conducted in an open manner, coordinated via the public discussion forum. The project is funded and maintained through voluntary contributions from the adopters.
As of 2020, the project is supported by several major organizations including NXP Semiconductors and the Dronecode Project.

Design

UAVCAN provides zero-cost abstractions that are approachable and familiar to software engineers without compromising on functional safety and determinism. As a new technology, it is unencumbered by legacy and borrows heavily from recent developments in the field of general information technology. The protocol offers a stateless publish-subscribe communication model where a node can begin operation immediately upon connection to the network to accommodate high-integrity applications.
The protocol has two clearly separated major components: the transport layer that works on top of reliable vehicular networks such as Ethernet or CAN FD, and the transport-agnostic presentation layer based on the so-called Data Structure Description Language. The protocol has been shown to be implementable in less than 1000 logical lines of code.
DSDL is ideologically similar to the interface description language used in ROS, except that it introduces additional static constraints in order to render the solution suitable for real-time high-integrity embedded systems. The similarity prompted some developers to interface ROS with UAVCAN using automated translation layers.

Core principles

The protocol is built around the following core design principles that are intended to ensure that the solution is well-suited for modern complex safety-critical vehicular systems.

UAVCAN/CAN

The CAN transport is built on top of CAN and CAN FD using 29-bit identifiers. The CAN payload includes a fixed-size overhead of one byte per frame for the needs of the transport layer.

UAVCAN/UDP

The UAVCAN/UDP transport has been proposed for real-time Ethernet-based vehicular networks. The design is said to be influenced by AFDX, DDS/RTPS, and SOME/IP.

Standard data types

Like other similar technologies, UAVCAN provides a library of common data types, managed and curated by the protocol maintainers, that are intended to address certain common issues in popular applications. These data types supplement vendor-specific or application-specific data types defined by adopters, much like a programming language would normally define a standard library to be relied upon by software developed by the user. The protocol specification provides a set of rules intended to avoid conflicts and enhance interoperability of data types defined by independent vendors.