自主机器人在线超分辨率致密环境建模理论及实验研究

Online environmental 3D dense mapping is one of key issues for the current autonomous robotics research. It can provide solid technical and theoretical support for the robots autonomous navigation ( e.g., online robot self-localization for GPS-denied cases, online robot local path planning) and the robot 3D structure reconstruction and inspection (e.g., building 3D reconstruction, bridge 3D inspection). This proposal investigates some key theoretical issues for autonomous robot online super-resolution dense 3D mapping by integrating a spinning 2D laser scanner, a high-resolution camera and low-cost IMU as the perception system. The spinning 2D laser scanner provides accurate 3D environmental measurements. While, the sampling rate is low and the 3D point-cloud collected by laser is sparse. The camera obtains dense environmental projection information, but no depth for each pixel. The IMU measures the robot state with very high sampling rate, but the IMU integral easily drift for navigation. As a result, combing the three types sensor for high-rate, accurate state estimation and super-resolution mapping is worth for exploration. The main feature of the proposal is twofold: online real-time mapping and super-resolution reconstruction. The key scientific issues and techniques of the research include three aspects: 1) Temporal-spatial fusion of sparse laser point-cloud and color image for super-resolution perception. 2) Inertia-aided 6D.O.F real-time odometry for laser 3D point-cloud Temporal-spatial registration; and 3) Inertia-visual SLAM for online global consistent dense 3D reconstruction. The scientific achievement of this research is to develop a online 3D super-resolution reconstruction theory for autonomous robotics. It can provide a solid technical basis for further research on robots long-term autonomous.

在线3D环境致密建模是当前自主机器人研究的核心议题之一，对于机器人自主导航（在线自定位、局部运动规划）、机器人3D结构检测（建筑物、桥梁检测）等领域有着重要实际意义。申报项目以“在线、超分辨率环境建模”为研究特色，目标在于时空融合稀疏激光感知和视觉致密感知，同时辅以惯性测量单元的状态感受信息，基于同时定位与地图构建（SLAM）理论对高质量致密3D环境模型在线重建理论与方法开展研究工作，以期综合利用激光感知的精确性、视觉感知的致密性、惯性信息的高数据率及短时平稳性。项目深入探讨激光-视觉时空融合的超分辨率致密环境感知、基于惯性信息辅助视觉里程计算的激光点云时空注册、基于关键帧的视觉-惯性紧耦合SLAM全局一致环境重建三方面关键科学问题，并通过样机集成与试验研究，探索机器人超分辨率致密环境建模的新技术，力争在自主机器人环境建模领域取得新突破。

项目面向智能移动机器人的自主定位与三维环境致密感知与构建展开研究，融合稀疏激光点云、高分辨率视觉感知图像以及高频的惯性信息，构建环境的致密点云3D模型，同时输出高频、高精度的定位信息，并构建多传感器环境感知的移动机器人试验样机平台完成关键技术的验证与应用。研究中通过视觉－惯性里程计算和激光里程计的外参手眼自动标定来解决多源异构感知数据的关联问题、并展开了多传感器数据紧耦合估计机器人位姿状态、惯性信息辅助激光点云时空注册、环境高精地图构建中的道路分割和动态障碍剔除等科学问题与关键技术的研究，并在机器人自主定位导航、机器人3D结构检测等领域对研究的科学问题展开应用，为进一步探索机器人智能自主行驶与在线三维致密地图构建奠定基础，因此具有重要的科学意义。研究成果在工业领域无人驾驶、桥梁检测、矿井探测、水下结构检测、灾难现场重建与搜寻等领域也具有广阔的应用场景。