Matt McDermott Comp 138: Reinforcement Learning
Part 1: Direct Friction Identification
cd final
<python viz2.py> - compares trajectories of "ground truth" and best estimated parameters
/tests/simplified_model/1link - graveyard of failed attempts at parameter estimation
Part 2: Torque control
cd torque_control
<python viz.py> #Runs OpenGL visualization of arm attempting to reach position setpoints from random initial configuration
# runs off of specified "path.npy" and "goal_path.npy" obtained from:
/tests/3DOF/gen_path_from_policy_variable_goal.py
/tests/3dof/
<python main_with_variable_goal.py> #trains model with random starting states, generates actor and critic checkpoints saved in
/checkpoints/ dir
/tests/3dof/
<python view_results.py> view actor loss, critic loss, and reward plots
/tests/3dof/
<python gen_path_from_policy.py> generates path of arm for viz basd on random inputs and the policy generated by the main func stored in the /checkpoints/ directory
/demo/ - videos and GIFs of different tests run on various policies
/tests/3DOF/ solve_EOM_combined_friction_model.py - solves EOM and saves as func to binary file