水果采摘机器人视觉定位大误差及其执行器使能容错机制

Uncertain factors made vision system occur large random errors easily, and it was difficult for the control system of picking robot to compensate these errors, which made picking disable. The design of traditional fruit-clamping mechanism often takes no account of how to correct these errors. In order to overcome these problems, the research on large random positioning errors of binocular vision and its transitivity and fault-tolerant theory of end-effector were put forward. Firstly, based on the experiment platform of vision-mechanism positioning, errors of image targets and reverse solving errors of pose position were investigated. The investigated contents include the errors caused by occlusion, false boundary and pixel registration and so on and their transitivity too. Then, Analyze the distribution and transfer characteristic of random errors, extract each factors that had great impact on the errors, establish an errors estimation model so as to provide a support for fault-tolerance design and make robots understand those uncertainty information easily. Besides, analyzed the clamping schemes of variety fruits and their biology common characteristics, study a novel fruit stem-clamping configuration which has certain universality and its fault-tolerance mechanism in every direction, build the three-dimensional transformation and mapping relationship between clamping and curvature of conversion cutting mechanism, establish a fault-tolerance model for co-operated work between clamping and cutting. Moreover, failure picking caused by random positioning error is regarded as a systematic “fault error”, explore the critical value, reliability and safety margin of the fault-tolerant system, optimize fault-tolerant design parameters, make the picking robot can finish errors correction and nondestructive picking under defects condition, realize precise picking. Finally, take litchi, citrus and grape as example, the fault-tolerance theory would be verified by self-developed patents and simulation platform.

不确定因素引起视觉产生了大的随机误差，控制难补偿它，使采摘失效，传统专用夹果机构不考虑对其修正。提出研究双目视觉定位大的随机误差与传递性及未端执行器使能容错机制。首先，在视觉与机构定位基础上，研究图像目标与位姿反求的误差，含遮挡、虚假边界、像素配准计算等综合误差及其传递性，分析随机误差空间分布范围、传递特点，分离出高频概率影响该误差的各因素，构建评估模型，为机器人理解不确定性信息和容错设计提供支持。然后，分析多种水果夹持与生物共性特征，研究有一定通用的感知夹持果枝的新构型及在各方向上对误差的容错机理，构建夹持和变切曲率与误差的三维变换和映射关系、夹切机构协调作业容错模型。进一步把失效作为系统缺陷，探索系统容错临界值、可靠性和安全裕度，优化机构容错设计参数，使机器人在缺陷条件下使能容错和无损采摘，实现精准作业。最后，以荔枝、柑桔等三种不同水果为例，用自有的发明专利和仿真平台验证理论的有效性。

针对采摘机器人野外作业环境的复杂性，创新提出双目视觉定位大的随机误差与传递性及未端执行器使能容错理论。首先，研究了目标图像与位姿反求算法、综合误差，构建评估模型，为机器人理解不确定性信息和容错提供支持。第二，分析了荔枝、葡萄、番石榴夹持与生物特征，研究有一定通用的夹枝与夹果的构型，构建了机构与视觉的误差关系、夹切机构与视觉协同作业的容错模型。第三，通过试验，分析失效原因，研究了避障算法，讨论视觉误差范围，给出了设计与控制的合理容错范围。把难以避免的外力造成的采摘失效作为系统缺陷，通过容错使机器人精准作业。项目研制了具有自主产权的6自由度轻量化番石榴采摘机器人，其负载3KG、整机24KG，视觉检测精度88%，成功率达75%，更换夹指可采摘荔枝，机器人具有网络传输功能。在相同负载、功能及精度情况下，解决轻量化机构、驱控设计、视觉定位难题，同比我们前期研发的荔枝采摘机器人，轻量化机器人整机的重量仅是前期的1/15.8。其视觉定位与控制、软件与算法、机械臂及末端执行机构、驱控一体的伺服驱动均具有自主产权，经鉴定达到国际先进水平。成果及理论可用于农业、民用与军用的视觉机器人野外作业。

内容获取失败，请点击重试