三维感知模型的训练及应用方法、装置、设备及机器人
By using high-precision scanning in real-world environments and reproducing robot motion in simulated environments, combined with parallel processing units to render light and determine the true semantic labels of voxels, the high-cost and low-efficiency 3D environment perception training problem in existing technologies is solved, achieving low-cost and high-efficiency 3D perception model training and stable environment perception.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SUZHOU GALBOT TECHNOLOGY CO LTD
- Filing Date
- 2026-04-27
- Publication Date
- 2026-07-17
AI Technical Summary
Existing technologies require the collection of a large amount of real machine data across the entire scene when training 3D environment perception models, resulting in high acquisition costs and low efficiency. Furthermore, they cannot effectively handle the posture changes and occlusion problems of non-rigid robots, leading to confusion of perceived targets and map instability.
By reconstructing the real environment through high-precision static scanning and combining it with the robot's real motion trajectory to reproduce the operating state in the simulation environment, high-precision simulation data is generated. Then, by rendering simulated light through a parallel processing unit, the true semantic labels of voxels are determined, thus achieving low-cost and efficient training data generation.
It reduces the cost of acquiring training data, improves the efficiency of data acquisition, ensures the stability and accuracy of 3D maps, avoids problems such as robot obstacle avoidance and target confusion, and adapts to the posture changes of non-rigid robots.
Smart Images
Figure CN122114046B_ABST