Quadruped-Experience-scripts

Python scripts to be implemented on the ODRI robot, inspired by script from the repositories sduminy/Inverse-Kinematics and sduminy/TSID

Description of the files

Inverse-Kinematics

All the files in this folder aims to create a simulation of walking with the robot.

initialization_simulation.py initialize the robot model (using robots_loader.py) and PyBullet simulator
walking_controller.py contains controller functions for the walk (the first one c_walking_IK() with a sinusoidal curves trajectory of the feet, the second one c_walking_IK_bezier() with Bezier curves as trajectory during the swing phase). Those controllers are based on Inverse-Kinematics : it computes the Jacobian of the errror to find the velocity which mminimizes the error.
PD.py is the PD torques controller. The controllers' output are torques, which are sent to the robot via PyBullet.
main.py computes the simulation in a for-loop, creates 2 numpy array variables Q and V to log the reference configurations and velocities and save them into a file reference_q_v.npz. They will serve as reference for the PD controller tested on the robot Inverse-Dynamics\walking_PD_controller.py.

Inverse-Dynamics

The files in this folder use Inverse-Dynamics to simulate/experience some movements on the robot.

Simulation

main_simu.py serves to test the different controllers on simulation first using PyBullet and the same environment as the one in Inverse-Kinematics
main_simu_12dof.py is for a simulation using TSID and contact forces, which force to use the 12 dof model of the robot
main_simu_walk.py was specifically design to test the walk in simulation (using the PD controller with the references logged in the reference_q_v_{slow/fast}.npz files

Experimentation

example.py is the example to test if the masterboard interface works well
masterboard_utils.py contains all functions from the masterboard (initialization of motors, saturation of the current, desired torques setting). Caution : the indexes of the configuration joints are not the same as the one used for the motors controlled by the masterboard. This file fix this problem.
libmaster-board-sdk.so and its Pyhton bindings libmaster-board-sdk_pywrap.so is the library of the masterboard
main_robot.py initializes the masterboard interface, calls a controller and computes the experimentation ; it also logs some interesting values into a logs.npz file using a class log_class.py. Those values can be displayed using plot_logs.py.
main_robot_12dof.py would be the same for a 12dof robot but it has not been tested so must need some adjustments
main_robot_walk.py is the main programm specifically designed for the experience of walking with the walking_PD_controller.py

Controllers

P_controller.py, PD_controller.py and PDff_controller are simple controllers to test first on the robot. They set a sinusoidal reference joint position to one or multiple joints. Those files were used to adjust the gains of the proportional and derivative terms.
Safety_controller.py is a pure derivative controller which is switched on when position bounds are reached.
EmergencyStop_controller.py is a zero-torques controller which is switched on when computation time is longer than 3 ms.
calibration_controller.py was used to calculate the real torques exerced by one joint (using a scale and the moment arm) to recover the relation between currents and torques torques_joints = Kred * torques_mot = Kred * Ki * currents_mot to see if it was correct.
walking_PD_controller.py is the controller used to experiment the walk on the robot by loading the reference joints positions and velocities with the reference_q_v_small.npz (for a trajectory period of 0.5 s) or reference_q_v_fast.npz (for a trajectory period of 0.3 s) files.
TSID_posture_controller.py uses TSID Joint Posture task (from all joints to zero radian, to the reference configuration straight_standing) ; was tested on the robot
TSID_foot_placement_controller.py uses TSID SE3Equality Task to set a foot to a desired position ; this has not been tested on the robot.
TSID_contacts_and_posture_controller.py and TSID_contacts_posture_and_com_controller.py were not tested on the robot. Those controller requires to load the 12 dof robot model to work.