Hand Game robot with Kotlin/Native and Raspberry Pi

Slide: https://speakerdeck.com/jinqian/bridge-the-physical-world-kotlin-native-on-raspberry-pi

Test the project

Circuit

This circuit schema represents what I use in my project:

Install C libraries on your Pi

This part explains how to install different dependencies on your raspberry pi in order to test the project.

Pre-requiste

Since we are using the Tensorflow 2.0, we will need glibc-2.28 which is only available starting from Debian 10 (Buster). So make sure your rasbian is upgraded, otherwise the tensorflow part will not be able to work.

Then make sure you have cmake installed on your raspberry pi:

$ sudo apt-get install cmake

lib pigpio

https://github.com/joan2937/pigpio

Connect to your pi with SSH:

$ git clone https://github.com/joan2937/pigpio
$ cd pigpio
$ make
$ sudo make install

libPCA9685

https://github.com/edlins/libPCA9685

$ git clone https://github.com/edlins/libPCA9685
$ cd libPCA9685 && mkdir build && cd build
$ cmake ..
$ make
$ ctest
$ sudo make install

Tensorflow

A japanese engineer has compiled the Tensorflow C libraries for raspberry pi and put it on github to share: https://github.com/PINTO0309/Tensorflow-bin

A huge arigato!!

$ git clone https://github.com/PINTO0309/Tensorflow-bin
$ cd Tensorflow-bin/C-library/2.0.0-armhf
$ ./install-buster.sh

Note: in order to use Tensorflow 2.0.0 you will need Raspbian/Debian Buster which is bundled with glibc 2.28.

/!\ For cross-compilation on a Linux machine or a MacOS, you have to download these libraries to your machine for linking during the compilation. In my code you can see they are gitignored because github does not allow file larger than 100MB. /!\

Generate the executable

Compile the project:

$ ./gradlew build

After a successful build you should be able to see that the binaries are generated under the build/bin/chifoumi folder. In order to test the binary, deploy it on your Pi via SSH, make sure you update the SSH configuration, source folder and destination folder in build.gradle with your own configurations:

$ ./gradlew deployOnPi

Note: you have to configure the passwordless SSH access on your Pi to make the SSH plugin work.

If it's the first time you deploy on your Pi, you need to make sure your .kexe file is executable, then execute the binary with sudo to test with your circuit:

$ chmod 777 chifoumi-robot.kexe
$ sudo ./chifoumi-robot.kexe

Other ideas

• Support linuxArm64?
• Desktop or other remote control

https://github.com/PINTO0309/Tensorflow-bin

Troubleshooting

Error when deploying with SSH plugin

If you ran into this problem:

* What went wrong:
Execution failed for task ':deployOnPi'.
> com.jcraft.jsch.JSchException: invalid privatekey: [B@57eea633

That is caused by the private key headers generated by different versions of OpenSSH. You can refer to this Stackoverflow post to solve the problem.

Reference

• Building Multiplatform Projects with Gradle: https://kotlinlang.org/docs/reference/building-mpp-with-gradle.html