Drone-based dynamic grasping of objects.
This project was tested with Python 3.10.12 on Ubuntu 22.04. To install, create a virtual environment and run:
TODO
You may now run the project by calling:
ipython -i simulate.py
Remember to click on the meshcat link!
In the debugging process I also created a ModelVisualizer script. You can use this to play with robot joints directly.
python visualize.py
TODO: put in diagram.
We split the drone and arm into two pieces which are effectively controlled independently. Each piece is composed of a controller and trajectory generator, which are separated in files Control.py and Traj.py. In the case of the drone:
DroneRotorController is the low-level controller. It implements a geometric controller based on this paper which essentially cancels the drone dynamics and uses proportional gains of position, velocity, rotation, and angular velocity to achieve a desired trajectory. Actuation is force commands.
FlatnessInverter is the trajectory generator. It takes position commands (instances of BezierTrajectory) and converts them into full pose commands using the differential flatness property of the drone platform.
The arm is a little different. Instead of pre-planning a trajectory using FlatnessInverter, the trajectory generator takes only a desired position and time. I pass a list of trajectory generators to the controller and it cycles through them.
JointController uses PI control to achieve a desired joint position with joint velocity commands.
ArmTrajectory is the trajectory generator, which takes position and time commands and plans a smooth trajectory to reach them.