Trajectory Optimization for Ropes Through Contact

Code and report will be available on my website soon:

Abstract:
Rope manipulation is challenging due to the infinite dimensionality of its configuration space. Planning with contact is challenging due to the discontinuous dynamics and rapid speed change. In this project, we are tackling both challenges in a task where a robot is moving a rope to a fully straightened and stable configuration. Since the robot has only one gripper, the task implicitly requires the robot to leverage available flat surfaces, such as a tabletop, to support and stabilize the rope. Our goal is to let the algorithm automatically discover the optimal contact-rich solution where the robot "chooses" to leverage the contact with the tabletop, rather than avoiding it. We modeled the rigid contact forces as impulse functions, approximated them using time-stepping averages, and applied trajectory optimization through contact to search for optimal trajectories. We have demonstrated the framework for a rope modeled as a chain with 10 rigid links in the x-z plane.

[rtedrake]

nice work, and a nice explanation. this was a very ambitious project, and you got quite far.

[tobiamarcucci]

Good job! Some time ago I also worked on the automatic parameter refinement idea, I didn't do it very seriously though. I'd be happy to know if you find anything interesting in that direction!