OIAC_whips

Background

The project aims to explore how to apply data-driven methods to realize the manipulation of deformable objects ,i.e., whips. The data-driven methods refer to Genetic Algorithm (GA), NLOPT, RL. In the real experiments, the vision tracking method is MeanShift, also comparing the performance with optic_flow and a DL python package called GOTURN. The following parts are introduced specifically.

Install

pip install -r requirements.txt

Usage

• Simulated Env

  • whip model
    The whip models are stored in /models folder
  • algorithms
    Each optimization method is sepeartely set in a main_xxx.py ('xxx': refers to the name of optimization methods). The file named main_noML.py without any optimization. Running these main files with simply command line: python main_xxx.py.

• Real Env

  • motor
    The motor part is explained in /dynamixel_motor_control_python folder. Note that, run main_kept.py to make the arm start with the same MuJoCo simulated position.
  • camera
    The project ustilizes IntelD435 to track the whip tip. /perception folder stores all vision related files, including tracking file and some useful tools. Track with command: python track.py. The function of useful tools can be easily understood by their names. The videos are captured from MuJoCo and real experiments.