AmyPhung / robo_dodgeball

This repo holds the code for the Introduction to Computational Robotics course final project

This repo holds code for the final course project in Computational Introduction to Robotics, FA 2020. The code is authored by Amy Phung, Everardo Gonzalez, and Nathan Faber.

Project Website: https://everardog.github.io/ml_comprobofinal/

Dependencies:

• inputs (for joystick controller) pip install inputs
• Tensorflow pip install tensorflow
  • If you get an error message about cuda when importing, run this: sudo apt install nvidia-cuda-toolkit

Usage

• To record dataset: roslaunch ml_comprobo record_training_data
  • Args:
    • dodgeball_prefix default: "ball"
    • robot_name default: "mobile_base"
    • num_dodgeballs default: "2"
    • use_joystick default: "true"
  • When using the joystick:
    • press a to start recording and start ball spawner
    • press b to stop and save dataset
    • use the left joystick to drive robot forwards/backwards
  • When using the keyboard:
    • press s to start recording and start ball spawner
    • press spacebar to stop and save dataset
    • press i or . to drive forwards/backwards
• To train dataset: python3 train_LTSM.py
  • Double-check the file to ensure paths are set correctly
• To test dataset in gazebo: roslaunch ml_comprobo run_model.launch
  • Make sure the model arg is set correctly in the launch file

TODO:

• make velocity relative to robot coords
• split training and data recorder
• Add param to auto-start ball spawn