Robotic Depowdering for Additive Manufacturing Via Pose Tracking

TL;DR

This paper addresses the depowdering bottleneck in powder-based additive manufacturing by introducing a vision-based robotic system that continuously tracks the 6D pose of powder-occluded parts and estimates depowdering progress. A key contribution is the Conditional Update ICP (CU-ICP), which couples an ICP-based template with a selective template-update strategy to maintain accurate pose estimates without large datasets or retraining. The system demonstrates real-time performance on a laptop CPU (up to 60 FPS) and robustly handles parts of varying shapes, both stationary and moving, enabling automated depowdering without pre-depowdering. Experiments show improved tracking accuracy and successful depowdering of diverse geometries, with practical implications for reducing manual labor and exposure to airborne powder.

Abstract

With the rapid development of powder-based additive manufacturing, depowdering, a process of removing unfused powder that covers 3D-printed parts, has become a major bottleneck to further improve its productiveness. Traditional manual depowdering is extremely time-consuming and costly, and some prior automated systems either require pre-depowdering or lack adaptability to different 3D-printed parts. To solve these problems, we introduce a robotic system that automatically removes unfused powder from the surface of 3D-printed parts. The key component is a visual perception system, which consists of a pose-tracking module that tracks the 6D pose of powder-occluded parts in real-time, and a progress estimation module that estimates the depowdering completion percentage. The tracking module can be run efficiently on a laptop CPU at up to 60 FPS. Experiments show that our depowdering system can remove unfused powder from the surface of various 3D-printed parts without causing any damage. To the best of our knowledge, this is one of the first vision-based robotic depowdering systems that adapt to parts with various shapes without the need for pre-depowdering.

Figures (12)

• Figure 1: (a) Examples of manual depowdering. Human operators use blowers, vacuums, and brushes to remove the unfused powder. (b) Overview of our depowdering system. The point cloud sequence from 3D cameras is fed into the visual perception system to track the 6D pose of the powder-occluded part. Based on the estimated pose, a depowdering path is generated for the robot to remove powder through vacuuming and air blasting.
• Figure 2: Example of a 3D-printed part losing balance during the depowdering process. As the vacuum constantly removes powder, the part gradually loses powder support and finally falls to one side.
• Figure 3: The architecture of our system. The inputs to the system are the CAD model and the point cloud can. When the depowdering progress is less than $\eta_1$, no tracking is performed, and the initial pose is used. Once the progress exceeds $\eta_1$, CU-ICP starts running. The robot first removes powder through vacuuming, and then finishes up depowdering by air blasting.
• Figure 4: A comparison of different template update strategies. (a) A propeller partially covered by powder. (b) The CAD model rendered according to the ground truth pose. (c) Pose-tracking result with strategy 1. (d) Template taken from the overlapping area between the CAD model and the scan. (e),(f) Pose tracking results with Strategy 2 and our strategy. With Strategies 1 and 2, the bottom of the propeller is mistakenly aligned to the powder surface, resulting in incorrect estimated pose. Our strategy resolves this issue.
• Figure 5: Visualization of template error accumulation with Strategy 2. The initial template aligns well with the point cloud (left). As tracking starts, although the target remains stationary, the sensor noise causes the oscillation of the estimated pose. The oscillation results in the template error accumulation, leading to an incorrect template and pose estimate.