DQN_flip_up_paper

Python 3.8
ROS melodic
Ubuntu 18.04
pytorch 1.10 CPU
robot hand setup :

https://kb.seedrobotics.com/doku.php?id=start

起動順番：

dynamixel サーバー起動

sudo chmod 777 /dev/ttyUSB0 roslaunch work controller_manager.launch roslaunch work start.launch All IO API are defined in DQN_flip_up_paper/dynamixel_driver/src/dynamixel_io.py
And parameter address are defined in DQN_flip_up_paper/dynamixel_driver/src/dynamixel_const.py

センサーのポート起動

python sensor.py

If raise errors of data size just rerun this file

キーボード入力

python teleop.py

no feedback

ロボットハンド制御システム

python mainCtrl.py main control method is defined here
the distance gap is calculated in main function
Action is received in ROS feedback
send hand motor information to datamerge

data merge

python DataMerge.py merge the sensor data and motor data
Input reward here each epoch

DQN training(When training)

python DQNTrain.py DQN is defined here and calculate the InputReward here

DQN testing(When testing)

python DQNTest.py need to input the trained model

When training

1.execute the above files
2.input 'w' in teleop(after DQN shows ">>>epoch XX")
3.input 's' in teleop when finish
4.input reward in datamerge
5.loop 1-4 until tarining process end

Rits-Interaction-Laboratory/DQN_flip_up_paper