Contact Map Generation for the Assessment of Robot Cleaning Tasks in Partially Unkown Environments

Kurzfassung

Solar energy is one crucial resource for planetary exploration. On Mars however, solar panels are prone to be coated by dust, which reduces efficiency or leads to complete power loss. On February 13th 2019, NASA announced the end of the Opportunity Mars rover mission, after almost 15 years. Since the rover had gotten into a Martian dust storm, NASA was not able to restore contact. The last message Opportunity send to Earth revealed, that it was very dark and that the on-board batteries were running low. This indicates, that the solar panels, which provide the rover with power, were heavily covered by dust and thus unable to collect enough sunlight. Equipping robots with the capability of maintaining solar panels, in order to prevent energy loss, will be a vital contribution to allow for successful Mars missions. Thus, future space robot assistants and rover will be able to create a sustainable pathway for human exploration, which includes the maintenance of solar farms. In this thesis a comprehensive approach is proposed to provide a humanoid robot with the skills for maintaining a solar panel farm. The robot is able to perform cleaning tasks on unknown surfaces, as well as creating a contact map of the cleaned area, which serves as action assessment indicator. In particular, a perception method is used to detect unknown objects by means of segmentation and primitive shape fitting. A motion planning framework is utilized to generate wiping motions w.r.t. geometric and semantic information of the surface and the used tool. A novel technique is introduced to facilitate assessment of the cleaning task by means of autonomously generated contact maps. The robot is able to determine the next solar panel, that requires maintenance, as the map is dynamically updated over time. This capability will enable future robots to autonomously maintain a solar panel farm.