One-Robot Wire Harness Installation without Offline Learning
Sequential Manipulation of Deformable Linear Object Networks with Endpoint Pose Measurements using Adaptive Model Predictive Control
Tyler Toner1,3, Vahidreza Molazedeh3, Miguel Saez3, Dawn M. Tilbury1,2, Kira Barton1,2
1 Department of Mechanical Engineering, University of Michigan
2 Robotics Department, University of Michigan
3 General Motors Research & Development
Preprint at arxiv.org/abs/2402.10372
Brief video summary
ICRA 24 final video submission twtoner out.mp4In this work, we study the manipulation of deformable linear object networks (DLONs): networks of connected, heterogeneous DLOs, a class of objects that includes automotive wire harnesses.
Wire harness installation requires mating all terminals to their receptacles, a task that remains a manual operation in automotive assembly due to the complex dynamics of DLONs.
While prior work has studied robotic DLO manipulation, we integrate practical constraints involving manipulation and perception in the development of our approach.
For a class of stiff DLOs (StDLOs), we demonstrate the existence of direct input-output dynamics, and find an approximate adaptive composite model that decomposes into a constant rigid body model and continuously updated local regression model.
As a result, we are able to use model predictive control to effectively control a DLON with a single manipulator while satisfying constraints.
See our supplementary video (above) and our manuscript for further details.