In discrete automated manufacturing, rigid vibratory feeders and mechanical jigs create crippling downtime and retooling overhead when handling high-mix, non-uniform industrial parts. SPACEX Industrial develops advanced 3D structured-light vision guidance systems that fuse geometric point-cloud filtering with deep-learning pose estimation, giving industrial robots adaptive spatial perception. Partner of HIPHOENIX (en.hiphoenix.com.cn).
1. Real-Time Point Cloud Denoising & Voxel Grid Downsampling
To eliminate optical artifacts on reflective metallic surfaces and ceramic coatings, our pipeline combines Statistical Outlier Removal (SOR) with adaptive bilateral smoothing, eliminating 99.8% of spatial noise. Voxel grid decimation cuts data density by 70% while safeguarding crucial edge geometries, yielding sub-120ms frame calculation latency.
2. 6-DoF Pose Coarse Estimation (SHOT) & Point-to-Plane ICP Refinement
Using Signatures of Histograms for Orientation and Topography (SHOT) descriptors, the algorithm identifies chaotic workpiece orientations in semi-structured bins. Subsequent Point-to-Plane Iterative Closest Point (ICP) refinement delivers grasping accuracies within ±0.3 mm and ±0.15 degrees.
3. Collision-Free Octree Trajectory Planning
Integrated with real-time octree spatial collision grids, our trajectory solver computes optimal robotic attack angles and retraction vectors within milliseconds, eliminating fixture impacts and guaranteeing uninterrupted 24/7 autonomous production.