ZCM: Zero Communications and Marshalling
ZCM is a micro-framework for message-passing and data-marshalling, designed originally for robotics systems where high-bandwidth and low-latency are critical and the variance in compute platforms is large.
ZCM is a publish/subscribe message-passing system with automatic message type-checking and serialization. It provides bindings for a variety of programming languages, and generates language-specific message serialization routines. Using a ZCM message feels natural in each language.
ZCM is transport-agnostic. There is no required built-in transport. Every transport is first-class. This is achieved by defining strict blocking and non-blocking transport APIs. As long as a transport implementation conforms to this API, it should work flawlessly with ZCM. This design allows ZCM to work well on anything from a high-end posix-based compute cluster with thousands of nodes to a low-end real-time embedded-system with no operating system.
ZCM is a derivation of the LCM project created in 2006 by the MIT DARPA Urban Challenge team. The core message-type system, publish/subscribe APIs, and basic tools are ported directly from LCM and remain about 95% compatible. While there are a handful of subtle differences between the two, the core distinguishing feature is ZCM's transport agnosticism. LCM is designed completely around UDP Multicast. This transport makes a lot of sense for LAN connected compute clusters (such the original 2006 MIT DGC Vechicle). However, there are many other applications that are interesting targets for ZCM messaging. These include: local system messaging (IPC), multi-threaded messaging (in-process), embedded-system peripherals (UART, I2C, etc), and web applications (Web Sockets). By refusing to make hard assumptions about the transport layer, ZCM opens the door to a wide set of use-cases that were neither possible nor practical with LCM.
To learn more about what ZCM tries to be, and its guiding principles, check out the Project Philosphy.
To dive, in and see some examples, check out the Tutorial.
If you have previously used LCM, check out From LCM to ZCM.
To learn how you can contribute to this project, check out Contributing
- Project Philosphy
- From LCM to ZCM
- ZCM Type System
- Transport Layer
- Embedded Applications
- Web Applications (coming soon)
- Dependencies & Building
- Frequently Asked Questions
- Continuous Integration
- Type-safe and version-safe message serialization
- A useful suite of tools for logging, log-playback, and real-time message inspection (spy)
- A wide set of optionally built-in transports including UDP Multicast, IPC, In-Process, Serial, and ZeroMQ
- A well-defined interface for building custom transports
- Strong support for embedded applications. The core embedded code is restricted to C89.
- Only one true dependency: A modern C++11 compiler for the non-embedded code.
Supported platforms and languages
- Web browsers supporting the Websocket API
- Any C89 capable embedded system
- C89 and greater
- MATLAB (using Java)
- Julia (both v1.6.0 and v0.6.4)
- Platform Support
- OS X
- Any POSIX-1.2001 system (e.g., Cygwin, Solaris, BSD, etc.)
- Consider porting the rest of the LCM languages
- Explore alternative messaging paradigms using ZCM Types (e.g. those found in ZeroMQ)
- Break from the original LCM APIs to improve API consistency
- Goal for v2.0
- v1.0 will always strive for API compatibility
Subtle differences to LCM
ZCM is approximately 95% API compatible with LCM. Porting existing Unix-based LCM
programs to ZCM is very easy in many cases. A quick
sed -i 's/lcm/zcm/g' works for
most applications. ZCM uses the same binary-compatible formats for UDP Multicast, Logging,
and ZCMType encodings. Thus LCM and ZCM applications can communicate flawlessly. This
allows LCM users to gradually migrate to ZCM.
zcm_get_fileno()is not supported
zcm_handle_timeout()is not supported
- Any applications using GLib via LCM may have build errors
- ZCM does not depend on GLib
- ZCMType drops support for the LCMType-style enums