In this thesis, we systematically analyze all major factors that affect a Camera-Space Manipulation (CSM) system’s positioning precision.
CSM is an effective control means to guide robots with machine vision in manipulation tasks. When using CSM, apart from error in establishing camera-space objectives, there are three major error sources that affect a robot’s end precision: camera optical model imperfection, robot kinematical model imperfection, and finite camera-image resolution. There are several factors that can mitigate the effect of the three major error sources and ensure sufficient precision, factors such as “preplan-trajectory” design (parameter initialization), robot end-tool cue configuration, and cue-data-weighting scheme. Prior to the work of this dissertation, our understanding of the three major error sources and mitigating factors, largely based on theoretical analysis and experimental conclusions, enabled us to achieve a sufficient level of precision for many manipulation tasks.
We seek herein to achieve much higher precision with CSM so that we can apply it to a much broader range of tasks. First, we set up a simulation platform to simulate a CSM system and investigate numerically the relationships among the major error sources, mitigating factors, and end precision. After that, we establish a high-precision, experimental CSM system to test the simulation’s prediction regarding all the conditions that culminate to a high level of precision. Based on this system, we further explore the quantitative impact on the end precision and the inter-relationships of the CSM measures, using both simulation results and experimental results. Eventually we can implement verified principles and effective measures into practice, and (ideally) realize robustly any desired level of end precision of a CSM system.
