基于机器视觉的搬运机器人研究综述 A Survey of Machine Vision for Material-Handling Robots
- To address perception bottlenecks as material-handling robots shift from teach-mode to autonomous agents, a full-stack evolution framework was led—“environment perception–path planning–precise manipulation–multi-robot coordination.” Surveying nearly 100 core domestic and international papers, the work compared classical geometric matching with deep learning (e.g., YOLOv13, Swin Transformer) across dimensions, establishing a multi-faceted evaluation benchmark for recognition and high-precision localization in unstructured scenes.
- For the trade-off between real-time obstacle avoidance and grasp accuracy in complex dynamic settings, visual SLAM, implicit 3D representations (SDF), and visuo-tactile fusion were analyzed for handling tasks. Quantitative runs on resource-constrained embedded platforms (e.g., Jetson) identified three core barriers—environmental adaptability, compute energy efficiency, and multi-sensor spatiotemporal alignment—clarifying optimization paths for lightweight perception in deployment.
Abstract
随着中小型仓储、实验室及家庭等场景对智能化搬运需求的攀升，机器视觉已成为搬运机器人从示教编程向自主智能体升级的核心驱动力。本文系统综述了机器视觉在搬运机器人领域的关键技术、研究进展与应用实践，并阐述目标识别定位、路径规划避障、抓取精度控制的技术演进，及多机器人协同的视觉信息共享机制，分析工业、仓储及特殊场景的落地价值。最后总结环境适应性、性能平衡等挑战，展望算法智能、感知多维等发展趋势，凸显机器视觉作为「感知中枢」的核心作用。