Technology Overview
AI Universal Picking: Auto-Labeled Data Generation Technology.
Traditional 3D point cloud analysis either requires the inputs of 3D CAD models or the assumptions about the shape of objects; however, these approaches are not applicable to identify various unknown objects. ITRI’s technology addresses the fundamental issue to prepare a large amount of labeled data for training an AI model. It can rapidly generate millions of labeled training data in a virtual world for the robot to learn the proper picking strategies. This technology minimizes both the manpower efforts originally wasted in data labelling, and costs.
Applications & Benefits
Users could quickly adopt the technology in robotic arm to pick new objects, rather than spend one to two weeks to adjust the picking algorithm to achieve the same result.
Auto-Labeled data generation and robust grasp point detection technologies framework.