David Martínez reports on his Master Thesis about Automatic Control and Robotics
Hello! I’m David Martínez, a Master’s degree student in Automatic Control and Robotics in the UPC. I joined the REEM@IRI team to do my master thesis in robotics. REEM@IRI is a collaboration between PAL Robotics , IRI and AESS to develop the skills needed by the humanoid service robot REEM – from the Barcelonese company PAL Robotics – to take part in the Robocup@HOME, competition, where robots have to solve everyday tasks, which include people recognition and speech interaction, robot navigation in a domestic environment, and object recognition and grasping.
Robot REEM from PAL Robotics, used in this project
For my master thesis, I worked on the open challenge of the Robocup@HOME where each team can demonstrate new and different skills. We chose to develop surface cleaning skills, which included erasing painted surfaces and cleaning small objects, like lentils or rice.
The method developed has 3 different parts: the perception to get a representation of the environment, a set of actions that are used to clean, and a planner to select the best action sequence for each state. REEM was equipped with a RGBD camera for the perception, and cleaning tools, like for example a cloth, to perform the actions successfully.
– To get observations of the environment, a RGBD camera is used, which provide both depth and color information. Depth is used to segment the dirty surfaces, while color is used to detect the dirty areas and represent them with a set of ellipses.
– We also defined a set of actions to clean, which consist in sequences of movements over the dirty areas to clean them. Each action gets better results with different types of dirty areas, making all actions useful for some kind of dirt.
Actions planned to clean a round drawing on the board
– We wanted to choose good action sequences to minimize the time spent cleaning, but we had uncertainty in the perceptions, and also stochastic actions. We decided to used PRADA planner, a fast probabilistic planner, to select good sequences for any given dirty surface. It improved significantly the results in complex tasks over using simple reactive methods.
The final result can be seen here:
REEM cleaning a whiteboard
It has been a very good experience being part of the REEM@IRI team. I have had the opportunity to learn a lot of things, worked with advanced real robots capable of performing many skills, met great people and I have had a very good time. I would like to thank everyone in the team, especially Dr. Ricardo Téllez from PAL Robotics, and Doctor Guillem Alenya from IRI who did a great job coordinating the efforts of the team, and I am looking forward to see what awesome skills we will make for future Robocup competitions.