动态性能优化的移动机器人视觉系统同时深度辨识与轨迹跟踪

The ability of external environment perception of mobile robot systems can be enhanced after combining vision sensors. By visual feedback, mobile robots can effectively accomplish tasks, such as tracking trajectories, pose stabilization, and grasping, so vision-based mobile robots are of great significance for both research and application. However, due to the nonholonomic motion constraint and unknown depth information, it is difficult to design high efficient control schemes for mobile robots, leading to poor dynamic performance, and it is also a challenge to realize online range identification while fulfilling visual control tasks simultaneously. Therefore, this project aims to propose a novel mobile robot visual trajectory tracking strategy with dynamic performance optimization, and simultaneously identifying depth information in the tracking process, thus the control effectiveness and perception ability could be improved. Specifically, the visual range observer is designed with terminal sliding mode at first to achieve finite-time convergence, and the projection function is introduced to dispose the visual target deviating from the camera field of view. Then, an online motion planning algorithm is designed for mobile robots in the existence of depth identification errors. Subsequently, a novel visual servo tracking control scheme is constructed under several uncertainties, which can tune the transient processes of the closed-loop system. Moreover, the robust robot dynamics controller is designed with input saturation avoidance. Finally, this project will be validated by abundant experiments.

将移动机器人与视觉传感器相结合，可提高机器人对外界环境的智能感知能力，继而利用视觉反馈可使机器人完成轨迹跟踪、定点控制和抓取等任务，因此该系统具有重要的研究与应用价值。受到非完整运动约束以及视觉信号深度缺失的限制，难以设计出高效可靠的控制方法，导致系统的动态性能较差；另外在实施视觉控制任务的同时，完成场景深度的在线辨识也具有挑战性。为此，本项目旨在提出一种可优化动态性能的移动机器人视觉轨迹跟踪控制策略，并在机器人运动过程中同时将场景深度准确稳定辨识出，从而提高系统的控制效果和环境感知能力。具体而言，首先利用终端滑模结构设计出有限时间收敛的视觉深度观测器，并设计投影函数以处理目标特征偏离视野情况；之后考虑深度辨识误差，设计移动机器人在线运动规划算法；继而设计暂态过程可调节的视觉跟踪控制器，并处理多种不确定因素；此后构造能避免输入饱和的机器人鲁棒动力学控制器；最后对所提方法进行大量的实验验证。

针对移动机器人的视觉伺服系统，项目中围绕视觉深度辨识、动态场景处理、自适应轨迹跟踪、系统参数标定等方面开展了研究，所取得的研究成果如下：. 1）针对视觉伺服任务中深度信息未知的情况，项目中根据并发学习框架利用系统历史和当前数据设计了未知参数自适应更新律，进而设计了移动机器人的视觉镇定控制律，经过稳定性分析得到了深度辨识误差和位姿镇定误差同时收敛的结果。2）针对动态场景的视觉伺服问题，在机器人工作空间中引入了监视摄像机，进而计算视觉目标的运动信息，实现在视觉目标发生移动时仍能使移动机器人运动至原期望位姿处。3）为解决无法预先获取期望图像的情况，提出了无需示教模式的机器人视觉镇定方法，根据视觉目标定义参考坐标系，辨识目标尺度后将机器人镇定至设定位姿处。4）针对未知的摄像机和移动机器人之间的平移外参数，设计了新颖的自适应非线性跟踪控制律，完成了平移外参数无标定的视觉伺服轨迹跟踪任务。5）提出了一种适用于轨迹跟踪任务的实用摄像机内参数自标定方法，针对内参数未知或不准确情况，在执行轨迹跟踪任务时快速完成内参数的标定。6）设计了一种同时优化摄像机外参数的视觉惯性里程计方法，恢复出旋转外参数、陀螺仪偏差、绝对尺度和重力矢量，在视觉惯导采样对齐阶段对平移外参数进行优化。. 项目按照计划书要求执行，发表SCI源刊论文6篇，包括IEEE Transactions论文4篇；授权国家发明专利1项，申请4项；参与国内外多个学术会议进行交流。