Welcome to flexrobotics, the home of roboflex.

Roboflex is a library for distributed robotics. Want to stream from a camera, run tensors through pytorch, integrate with other sensor streams, and control actuators and robots? Roboflex is for you.

Roboflex is:

• Simple to install: pip install roboflex, then pip install roboflex.realsense etc.
• Simple in concept: distribute Nodes, which receive and publish Messages over pub-sub networks. Threading model is straightforward.
• Simple to implement: use python to create and run Node graphs. Roboflex requires minimal code, and no non-code artifacts (no yaml, no xml).*
• Uses dynamic, not static, message types; more like json than protobuf. Roboflex uses flexbuffers as the message protocol.
• Performant: roboflex is written in c++. Flexbuffer message protocol supports 0-copy serialization into xtensor, eigen, and numpy.
• A library, not a framework: you control main().

The core idea of roboflex is the 'computation graph'. Nodes send Messages to other nodes, over any transport. You configure this computation graph in a small python script. See the 'core' documentation for more info: (https://github.com/flexrobotics/roboflex).

Roboflex is structured as a single core library (roboflex), in the roboflex repository. It supports no physical devices itself. Other repositories support devices, actuators, utilities, etc. This is the current set of repositories in roboflex; each one is named "roboflex_something". If you would like to support other devices or functionality, you can name your repo anything you'd like.

• roboflex (core) (https://github.com/flexrobotics/roboflex)

  Definitions and imlpementations of Nodes, RunnableNodes, Messages, python wrappings, and various serialization utilities. Also contains a small library of utility nodes:

  • CallbackFun
  • FilterFun
  • FrequencyGenerator
  • LastOne
  • MapFun
  • MessagePrinter
  • GraphRoot

• roboflex_examples (https://github.com/flexrobotics/roboflex_examples)

  Simple examples of roboflex individual components are in those components' examples/ directory. Examples that use two or more components, or are more complex, are here.

  • realsense_tv.py: television for the Intel Realsense camera.
  • webcam_tv.py: television for USB-UVC compatible web cameras.
  • audio_tv.py: displays an audio feed in real-time.
  • camera_follow: A pan-tilt dynamixel-driven camera mount that follows your face in realtime.

Nodes that broadcast data read from devices, such as images, audio streams, etc.

• roboflex.audio_alsa (https://github.com/flexrobotics/roboflex_audio_alsa)

  Audio-broadcasting node that uses linux's Advanced Linux Sound Architecture.

• roboflex.audio_sdl (https://github.com/flexrobotics/roboflex_audio_sdl)

  Audio-broadcasting node that uses Simple Directmedia Layer.

• roboflex.realsense (https://github.com/flexrobotics/roboflex_realsense)

  RealsenseSensor node, a driver for the Intel Realsense rgb/depth camera.

• roboflex.webcamuvc (https://github.com/flexrobotics/roboflex_webcamuvc)

  WebcamSensor node, a driver for USB UVC-compatible web cameras.

• roboflex.dvs (https://github.com/flexrobotics/roboflex_dvs)

  DVSSensor node and others, a driver for a particular model of Dynamic Vision Sensor (otherwise known as an 'event camera').

• roboflex.visualization (https://github.com/flexrobotics/roboflex_visualization)

  RGBImageTV and DepthTV Nodes, that can visualize rgb and depth camera streams, so far.

• roboflex.imgui (https://github.com/flexrobotics/roboflex_imgui)

  Visualizers using IMGUI and IMPLOT.

• roboflex.dynamixel (https://github.com/flexrobotics/roboflex_dynamixel)

  Support for dynamixel motors.

• roboflex.transport.zmq (https://github.com/flexrobotics/roboflex_transport_zmq)

  Zero-mq support for pub-sub.

• roboflex.transport.mqtt (https://github.com/flexrobotics/roboflex_transport_mqtt)

  MQTT support for pub-sub.

• roboflex.util.jpeg (https://github.com/flexrobotics/roboflex_util_jpeg)

  Converts (C,H,W) uint8 rgb image tensors to jpegs, in memory and file, and back.

• roboflex.util.png (https://github.com/flexrobotics/roboflex_util_png)

  Converts (C,H,W) uint8 rgb image tensors to pngs, in memory and file, and back.

• In python, roboflex is very simple to install and use. In c++, Roboflex currently uses cmake to build. Yeah, it's a mess. Open to ideas.