Sequential Manipulation of Deformable Linear Object Networks with Endpoint Pose Measurements using Adaptive Model Predictive Control

TL;DR

This work tackles single-arm planar manipulation of DLONs using only terminal pose measurements, addressing unknown, time-varying DLON dynamics. It reveals that output dynamics can be approximated by a polynomial model that decomposes into rigid-body motion plus residual terms, enabling an adaptive composite model for control. The authors develop an adaptive model predictive controller (MPC) with online model adaptation and a simple sequence planner to perform sequential terminal manipulations, validating the approach in both simulation and physical harness experiments. The results show robust, constraint-aware endpoint control without full state estimation, with practical implications for automotive harness installation and other DLON manipulation tasks. Limitations include a focus on planar, small DLONs and the need for state-estimation extensions and non-planar generalization in future work.

Abstract

Robotic manipulation of deformable linear objects (DLOs) is an active area of research, though emerging applications, like automotive wire harness installation, introduce constraints that have not been considered in prior work. Confined workspaces and limited visibility complicate prior assumptions of multi-robot manipulation and direct measurement of DLO configuration (state). This work focuses on single-arm manipulation of stiff DLOs (StDLOs) connected to form a DLO network (DLON), for which the measurements (output) are the endpoint poses of the DLON, which are subject to unknown dynamics during manipulation. To demonstrate feasibility of output-based control without state estimation, direct input-output dynamics are shown to exist by training neural network models on simulated trajectories. Output dynamics are then approximated with polynomials and found to contain well-known rigid body dynamics terms. A composite model consisting of a rigid body model and an online data-driven residual is developed, which predicts output dynamics more accurately than either model alone, and without prior experience with the system. An adaptive model predictive controller is developed with the composite model for DLON manipulation, which completes DLON installation tasks, both in simulation and with a physical automotive wire harness.

Figures (4)

• Figure 1: Framework for robotic installation of deformable linear object networks (DLONs), e.g., automotive wire harnesses, using only terminal pose measurements, or outputs. The controller leverages a composite input-output model containing a DLON-agnostic rigid body model and a DLON-specific local model learned in situ without prior learning or online excitation.
• Figure 2: Representative pose trajectory $\rho_{t,k} = [x_{t,k} \, \, y_{t,k} \, \, \theta_{t,k}]^\top$ of one free terminal $t$, selected from $\mathcal{D}$ (solid lines). The reconstruction of $\rho_{t,k}$ using the output map network and its inverse network, $\hat{h}(\hat{H}(\cdot))$ (dashed lines) is close enough to the original to be mostly obscured. The prediction of $\rho_{t,k}$ from $\rho_{t,0}$ using the polynomial model $f^{\boldsymbol{y}}_{s}$ (dotted lines) tracks the original initially, but deviates in $\theta$ after several seconds.
• Figure 3: Simulated DLON experiments. Each terminal must be brought to the static receptacle of the same color without leaving the yellow workspace. In (a)-(b), the only obstacles are the receptacles themselves. In (c), the DLON must fit through a narrowing created by obstacles (white). In (d), the DLON must navigate around a blocking wall of obstacles.
• Figure 4: (a) Physical setup with 6-DOF manipulator and modified 3-terminal automotive wire harness. (b-e) Physical experiments with harness configurations in increasing order of challenge. In (b)-(d), terminals are initialized away from obstacles. In (e), terminal $t_2$ is initialized near a receptacle, preventing grasping; the robot grasps and uses it to pull the violating terminal away from the obstacle before continuing installation.