Session: CIE-05-01CIE Graduate Student Poster Symposium

Paper Number: 74869

Start Time: August 18, 10:00 AM

74869 - A Mobile Manipulator System for Accurate and Efficient Spraying on Large Surfaces 

The ability to spray on large surfaces is essential in many applications. This task can be ergonomically challenging. Currently, fixed robotic manipulators are being used for this purpose. However, a fixed manipulator may not be able to spray on the entire surface due to limited reachability. We present a mobile manipulator system that can accurately and efficiently spray on large surfaces by automatically generating a plan based on the given spraying task. Each spray tool behaves differently. To correctly position and orient it with respect to the mural and determine its velocity, we need to develop a model of the spray process. Not using the right process parameters will yield significant errors. We use self-supervised learning to build a process model that describes how process parameters affect the spray performance. This process model is used to generate spray tool trajectories. We demonstrate the use of self-supervised batch learning to reduce the number of experiments needed to create a model of the spray tool. The next step is to generate mobile manipulator motion plans that can move the spray tool along the desired trajectories. This requires us to compute mobile base placements as well as robot arm trajectories while meeting reachability and velocity constraints. We are interested in minimizing the number of base placements to improve task efficiency. We report a mobile base placement planner that determines the minimum base locations required to carry out the spraying task. When the robot begins to execute the robot motion plan, we may find that the spray width has some variance. We need to use perception to characterize the width and quality of the spray. This characterization allows us to either update the spray tool model or estimate the error in the executed tool-path. We have developed an image-based perception pipeline that enables the robot to characterize spraying error. These algorithms have been experimentally verified in this paper by having a mobile manipulator spray paint a large mural. We demonstrate that our approaches enable a mobile manipulator to spray accurately and efficiently on large surfaces.

 

Presenting Author: Neel Dhanaraj Center for Advanced Manufacturing

Authors:

Neel Dhanaraj Center for Advanced Manufacturing