Improving Manipulation and Control of Search and Rescue UGVs Operating Across Autonomy Levels

Sammanfattning: Robots are often used for “dirty, dull and dangerous” jobs, where time, money or lives can be saved. A field with dangerous situations is search and rescue, with structural collapses and toxic environment. In those situations, robots have the potential to save lives.In this thesis we discuss improvements to unmanned ground vehicles (UGVs) for search and rescue in terms of manipulation and control. From the use of developments in video games for teleoperation, presentation of signal strength prediction, to ideas from 3D CAD for inspection tasks, and finally formation keeping for mobile manipulators. The problems will be addressed at different autonomy levels, regarding the robot operator interaction.We first consider teleoperation of an unmanned ground vehicle. Through a user study we compare two control modes, the traditional Tank Control and the video game inspired Free Look Control. Then, knowing how important connectivity is for teleoperation, where low connectivity can lead to the robot being abandoned, we propose a user interface which combines predicted radio signal strength with Free Look Control.Next we consider teleoperation of a mobile manipulator, which allows for inspection tasks. This adds complexity to the operator in terms of control, where not only the platform itself, but also the manipulator, has to be controlled. Inspired by 3D CAD software, where a core functionality is inspection of an object, we propose the control method Orbit Control. The system can assist with controlling some degrees of freedom of the robot, while the operator focuses on the final inspection and interaction.Finally, we consider a group of mobile manipulators transporting a larger object in a high obstacle density environment. A partially collapsed structure can contain areas which are not suitable to move in. To find a path for navigation, moving information while avoiding obstacles, can be challenging. We propose a combination of rapidly-exploring random tree (RRT) and constraint based programming, leading to a more efficient approach at high obstacle densities.