Particle Filters in Latent Space for Robust Deformable Linear Object Tracking


Tracking of deformable linear objects (DLOs) is important for many robotic applications. However, achieving robust and accurate tracking is challenging due to the lack of distinctive features or appearance, the object’s high-dimensional state space, and the presence of  occlusion. In this letter, we propose a method for tracking the state of a DLO by applying a particle filter approach within a learned lower-dimensional state embedding. We use an autoencoder to learn a latent space embedding for DLO states. The dimensionality reduction preserves state variation, while simultaneously enabling a particle filter to accurately track state evolution with a practically feasible number of particles. Compared to previous works, our method requires neither running a high-fidelity physics simulation, nor manual designs of constraints and regularization. Our method can be initialized without state information and results in accurate tracking even under complex DLO motions and in the presence of severe occlusions.

Code and examples: