Workspace Registration and Collision Detection for Industrial Robotics Applications

TL;DR

The paper addresses the need for accurate workspace registration and collision awareness in industrial robotics by comparing sensors and detailing a pipeline from sensing to a collision-environment representation. It combines point-cloud preprocessing and region-growing segmentation to produce cuboid-based environment approximations, then evaluates two collision-detection strategies: SAT with simple spherical approximations and Minkowski-difference based detection using GJK (with EPA for depth). The main contributions include a complete sensing-to-encoding workflow, analysis of bounding-box alignment to preserve usable space, and a comparative study showing that Minkowski-based collision detection reduces constraint count in planning. The work highlights practical implications for safer, more efficient collision-free motion planning in complex production environments.

Abstract

Motion planning for robotic manipulators relies on precise knowledge of the environment in order to be able to define restricted areas and to take collision objects into account. To capture the workspace, point clouds of the environment are acquired using various sensors. The collision objects are identified by region growing segmentation and VCCS algorithm. Subsequently the point clusters are approximated. The aim of the present paper is to compare different sensors, to illustrate the process from detection to the finished collision environment and to detect collisions between the robot and this environment.

Figures (7)

• Figure 1: Comparison of sensors
• Figure 2: Comparison of sensors
• Figure 3: Clustered example scenery
• Figure 4: PC with axis aligned and object aligned bounding boxes
• Figure 5: Approximation of the robot environment Fig. \ref{['fig:workspace']}