Skip to content

Latest commit

 

History

History
15 lines (13 loc) · 1.89 KB

README.md

File metadata and controls

15 lines (13 loc) · 1.89 KB
Raw

CACTO: Continuous Actor-Critic algorithm with Trajectory Optimization

  • main implements CACTO with state = [x,t]. Inputs: test-n (default: 0), system-id (default:'-'), seed (default: None), recover-training-flag (default: False), nb-cpus (default: 15), and w-S (default: 0).
  • TO implements the TO problem of the selected system whose end effector has to reach a target state while avoiding an obstacle and ensuring low control effort. The TO problem is modelled in CasADi and solved with ipopt.
  • RL implements the acotr-critic RL problem of the selected system whose end effector has to reach a target state while avoiding an obstacle and ensuring low control. It creates the state trajectory and controls to initialize TO.
  • NeuralNetwork contains the functions to create the NN-models and to compute the quantities needed to update them.
  • environment contains the functions of the selected system (reset, step, and get-end-effector-position functions).
  • environment_TO contains the functions of the selected system implemented with CasADi (step, and get-end-effector-position functions).
  • replay_buffer implements a reply buffer where to store and sample transitions. It implements also a prioritized version of the replay buffer using a segment tree structure implemented in segment_tree to efficiently calculate the cumulative probability needed to sample.
  • robot_utils implements the dynamics of the selected system with Pinocchio.
  • plot contains the plot functions
  • system_conf configures the training for the selected system.
  • urdf contains system URDF file (double integrator and manipulator).

Systems: single integrator (system-id: single_integrator), double integrator (system-id: double_integrator), car (system-id: 'car'), car_park (system-id: 'car_park'), and 3 DOF planar manipulator (system-id: manipulator)