PaintNet: 3D Learning of Pose Paths Generators for Robotic Spray Painting

Abstract
Optimization and planning methods for tasks involving 3D objects often rely on prior knowledge and ad-hoc heuristics. In this work, we target learningased longorizon path generation by leveraging recent advances in 3D deep learning. We present PaintNet, the first dataset for learning robotic spray painting of free-form 3D objects. PaintNet includes more than 800 object meshes and the associated painting strokes collected in a real industrial setting. We then introduce a novel 3D deep learning method to tackle this task and operate on unstructured input spaces—point clouds—and mix-structured output spaces—unordered sets of painting strokes. Our extensive experimental analysis demonstrates the capabilities of our method to predict smooth output strokes that cover up to 95% of previously unseen object surfaces, with respect to ground-truth paint coverage.


Authored by Gabriele Tiboni, Raffaello Camoriano, and Tatiana Tammasi from Politecnico di Torino. This work was supported by the EFORT group, providing the authors with domain knowledge, original object meshes, trajectory data, and access to the proprietary spray painting simulator used during the experiments.

Dataset

We introduce the PaintNet dataset to accelerate research on supervised learning for multiple pose paths prediction on free-form 3D objects. PaintNet includes more than 800 object meshes and the associated spray painting strokes collected in a real industrial setting. The data currently covers four object categories of growing complexity: cuboids, windows, shelves, containers. All object meshes are already provided in a subdivided, smoothed watertight version to avoid sharp edges and holes. For each object, the associated unordered set of spray painting paths (a.k.a. strokes) is given, each being a sequence of end-effector poses. The 6-dimensional poses encode the 3D position of the ideal paint deposit point—12cm away from the gun nozzle—and the gun orientations as Euler angles. Each pose is collected by sampling from the end-effector kinematics at a rate of 4ms during offline program execution.

Download

Coming Soon

Results

Coming Soon

Citing

Coming Soon