Probabilistic Pose Estimation of Deformable Linear Objects

Abstract

This paper presents a probabilistic framework for online tracking of nodes along deformable linear objects. The proposed framework does not require an a-priori model; instead, a Bayesian Committee Machine, starting as a tabula rasa, accumulates knowledge over time. The key benefits of this approach are a lack of reliance upon extensive pre-training data, which can be difficult to obtain in sufficiently large quantities, and the ability for robust estimation of nodes subject to occlusion. Another benefit is that the uncertainties obtained during inference from the underlying Gaussian Processes can be beneficial towards subsequent handling tasks. Comparisons of the non-time series framework were conducted against conventional regression models to measure the efficacy of the proposed framework.

Particle Filters in Latent Space for Robust Deformable Linear Object Tracking

“…In general, the high dimensional state poses a challenge for tracking while tracking in a lower dimensional representation from principal component analysis or Gaussian process latent variable models [32] is a feasible solution. Instead of employing a data-driven dimension reduction, Lai et al [33] rely on an assumption of a planar scenario to transform the DLO to a sequence of angular offsets. However, their strategy is not feasible for a 3D scenario.…”

Heuristic Detection of Recovery Progress using Robotic Data

“…Factors such as complexity of the data analysis and the metrics used to gauge process are rarely discussed [11]. A popular method to approach complexity reduction is the use of machine learning [12]. Using machine learning to build a model of a user's limb capability facilitates the potential to detect a change in user limb capability.…”

Measuring Bending Angle and Hallucinating Shape of Elongated Deformable Objects

“…In the literature, there are several ways to represent the geometric shape of the DLO. The most straightforward one is to represent it as a sequence of points [10], [11]. However, more complex models are usually necessary for accurate cable modeling and tracking, like a B-spline model with multiple chained random matrices, proposed in [6].…”

“…While there are attempts to use data from tactile sensors [16], the most successful way to perceive the DLO shape is to use vision and depth sensors. One of the most straightforward approaches to DLO shape tracking is to use the fiducial markers located along the DLO, and track them [11] or use them to estimate the shape of a DLO [17]. A similar approach was presented in [18], where colors denote consecutive rope segments.…”