robotic-powder-weighting

Getting started

To build environments

• install pytorch (support 1.12.1)
• install isaacgym
• install ROS (for real-robot)

Experiments

check behavior of the powder weighing simulator

This code shows powder shaking actions. You can control the environment by joystick controller.

cd simulation
python test_env_hellow.py --render

test powder weighing

This code shows powder shaking actions. You can control the environment by joystick controller.

cd simulation
python test_env_stick.py --render

learning policies

This code shows how to run the learning policies. Particularly, isaacgym have a memory-reek error. So, in this experiments the environment is reset by subprocess.

cd simulation
python root_learning.py --flag 1111111111 --render

flag means what domain parameters are fixed on learning and action definitions. 1 means random or set, 0 means fixed or removed. All parameters are bellow.

• incline_flag
• shake_flag
• ball_radius_flag
• ball_mass_flag
• ball_friction_flag
• ball_layer_num_flag
• spoon_friction_flag
• goal_powder_amount_flag
• shake_speed_weight_flag
• gravity_flag

render means learning with rendering the environment or not.

test policies on real robot

1. build up ros packages in catkin_ws
2. set up weighing machine
   • connect data cable from PC to the machine.
   • check USB port name(ex: /dev/ttyUSB or COM2)
   • set Format to 5 of weighing machine
   • set unit to mg of weighing machine
   • sudo chmod a+rw /dev/ttyUSB0
3. run the robot with learned policies

cd real_robot/powder_task
python PowderEnvironmentController.py --test --dir ./policy_directly/Policy.pth