The Dextra rock-scissor-paper robot perception pipeline redone in python using pyaer and tensorflow. See also

• The Dextra tendon driven hand paper and design https://sensorsini.github.io/dextra-robot-hand/
• The Arduino firmware https://github.com/SensorsINI/DextraRoshamboHand

The pretrained network is a 16-bit quantized weight and state CNN.

Requirements

• OS: Fully tested on Ubuntu 18.04 and 22.04

• Python 3.9 Important, necessary for Tensorflow 2.5.0 required to run the roshambo CNN

• Tensorflow 2.5.0

• CUDA 10.2+ or whatever comes with tensorflow install

  • Keras: 2.5.0
  • pyaer https://github.com/duguyue100/pyaer

• sensors DAVIS camera

• Robot hand with Arduino control via USB serial port bytes if you want to see the hand move; see project https://github.com/SensorsINI/DextraRoshamboHand

Setup

Project includes pycharm .idea/ folder and vscode .vscode/ folder.

System level prerequisites

1. Install libcaer. Can be installed with sudo apt-get install libcaer-dev, otherewise see https://gitlab.com/inivation/dv/libcaer.

Make a conda environment

Create the environment, activate it, then in it install the libraries. We recommend you use conda because it will download the necessary python version 3.9. (That is the last python version to have tensorflow 2.5.0 which this project uses.)

conda create -n roshambo python=3.9
conda activate roshambo

Other requirements

See requirements.txt for libraries needed. Install them to the new conda env from the conda prompt with

conda activate roshambo # probably already activate
pip install -r requirements.txt

If you have trouble with pyaer, see https://github.com/duguyue100/pyaer. It should work for linux and mac OS intel silicon machines. Windows probably will not work natively, but you can run the code in a WSL2 Ubuntu virtual machine using https://github.com/dorssel/usbipd-win to map the USB port to WSL2 Ubuntu.

USB/pyaer on Windows

Recommand to use WSL2 and the vscode plugin https://marketplace.visualstudio.com/items?itemName=thecreativedodo.usbip-connect to bind the USB port to WSL2 Ubuntu VM

Running Dextra

Run roshambo; it uses multiprocessing to launch 2 subprocessees, producer and consumer. (You can run these separately for testing.)

Run two processes, producer and consumer.

1. connect hardware: DAVIS to USB and Arduino to USB.
2. Find out which serial port device the hand controller Arduino appears on. You can use dmesg on linux. You can put the serial port into globals_and_utils.py to avoid adding as argument.
3. In terminal run producer

python -m roshambo