ZuPi - Raspberry Pi on Zumo Robot

END OF LIFE

The ZuPi project is leveraging robotalks for zero-coding implementation.

WORK IN PROGRESS

This repository contains the source code for ZuPi projects - the robots built with Raspberry Pi (or compatible alternatives) on Zumo robot.

NOTE: a lot of in-progress/early-stage work, lack of docs, tests, likely to break!

Looking for documents about the details, including hardware and software? Please visit the gitbook about this project ZuPi Build Memo.

Getting Started

Linux is the only supported platform, however you can build from Mac/Windows, thanks to HyperMake.

To build everything:

hmake build-*

To build individual targets:

hmake build-zpi1-brain-*   # build zpi1 brain for multiple archs
hmake build-zpi1-vision-*  # build zpi1 compute vision module for multiple archs
hmake pack-zpi1-robot      # create Node.js package for robot logic (tbus based)
hmake build-smartcam1d-*   # build smartcam1d algorithm simulation

Deploy And Play

Install the dependencies:

• Node.js version 4.5 or above
• v4l2rtspserver in PATH
• Enable camera and I2C (for servo controller)

Deploy robot package

hmake pack-zpi1-robot
scp out/zpi1-robot-0.0.1.tgz pi@robot:/root/
scp out/armhf/bin/zpi1-brain pi@robot:/usr/local/bin/
ssh pi@robot
    tar zxf zpi1-robot-0.0.1.tgz
    cd package
    npm install  # this takes a while
    DEBUG=* zpi1-brain conn -- node app.js connect '{{.host}}' -p '{{.port}}'

Now the robot is looking for brain (the controlling logic)...

On the controller (laptop or a server), build the brain

hmake build-zpi1-brain-amd64 build-zpi1-vision-amd64
export PATH=$PATH:`pwd`/out/amd64/bin
LOGXI=*=DBG zpi1-brain brain

When the brain is connected, it will start controlling the robot... and enjoy!

Simulation

Folder sim/smartcam1d contains the simulation code to experiment the controlling logic of moving the camera. It's easy to add/modify algorithms (see algo_*.* files).

To build the code

hmake build-smartcam1d-amd64

You need sim-ng (see Simulator) installed in PATH:

sim-ng vis -I sim/visualizer -- out/amd64/smartcam1d sim -l fido

Point your browser to http://localhost:3500, and you will see the object and camera. The reinforcement learning algorithm tries to move the camera to center the object. Once it centered the object, click on the screen to replace the object and let the algorithm center the object again. And you will see how fast the algorithm converges, or maybe never...

References

The project is based on technologies of

• ThingsBus
• Simulator