Object Isolation In A Random Environment Using Manipulation Primitives Approach

• Main Author: Quah, Jit Shen
• Format: Thesis
• Language: English; English
• Published: 2019
• Subjects: T Technology (General); TA Engineering (General). Civil engineering (General)
• Online Access: http://eprints.utem.edu.my/id/eprint/24617/1/Object%20Isolation%20In%20A%20Random%20Environment%20Using%20Manipulation%20Primitives%20Approach.pdf; http://eprints.utem.edu.my/id/eprint/24617/2/Object%20Isolation%20In%20A%20Random%20Environment%20Using%20Manipulation%20Primitives%20Approach.pdf; http://eprints.utem.edu.my/id/eprint/24617/; https://plh.utem.edu.my/cgi-bin/koha/opac-detail.pl?biblionumber=117087
• Institution: Universiti Teknikal Malaysia Melaka

As autonomous robots become less task specific to able to handle a larger variety of task that it may come across in the world, its ability to isolate the environmental objects from the object of interest becomes an increasingly important ability to have. This is in comparison to the current literature interest which focuses on the isolation of the object from the environment due to its precise nature. This method often consists of two distinct actions which is the action to improve the robot’s perception followed by another to manipulate the object of interest. This however becomes a problem when considering it might be impossible or undesirable to manipulate the object of interest such as the case of cleaning up archeological finds or even rescue missions of earthquake victims. In cases such as those, the precise nature of the actuation becomes a hindrance due to the long duration needed in order to manipulation every single object in the field one object at a time. This research explores the idea of combining both actions to improve perception as well as action to manipulate the object into one single motion with the added goal of manipulating as many objects as possible in a single action with minimal/no effect towards the object of interest. This research proposes three novel algorithms on a robot to plot out the position for each unwanted objects and its destined position as well as its trajectory and then utilizes manipulation primitives (pushing motion) to move said object along the planned trajectory. The algorithm was demonstrated using a KUKA Youbot with a camera fitted above the workspace. Results from the experiment indicated that all proposed algorithms successfully reduced the number of manipulations per object up to 0.605 manipulations albeit with a small tradeoff in accuracy from 97.7% to 93% which translates to average of 0.85cm/actuation for MSMAPPS, 0.75cm/actuation for MSMAPOS and finally 0.27cm/actuation for BSMAPOS. These results indicate a relatively small displacement per actuation at 4.35% displacement per actuation, 3.75% displacement per actuation, and 1.35% displacement per actuation relative to the workspace respectively. As a conclusion, the proposed methods is shown to be significantly more efficient then current methods employed in the field of object isolation in exchange for a small reduction in performance in terms of accuracy of actuation.