“I Want to Give Eyes to the CoBot Robot”
CoBot is a robot carried out by the research team of Manuela Veloso, at Carnegie Mellon University. The goal of this team is to contribute to a multi-robot, multi-human symbiotic relationship, in which robots and humans coordinate and cooperate as a function of their limitations and strength.
Susana Brandão, a dual degree doctoral student in Electrical and Computer Engineering (ECE), co-advised by Manuela Veloso at CMU and João Paulo Costeira, from Instituto Superior Técnico da Universidade Técnica de Lisboa (IST/UTL), is working with this robot and believes that she will help CoBot to “see” in the future.
Susana Brandão started her Ph.D. in 2009/2010 in Portugal, at Instituto Superior Técnico da Universidade Técnica da Lisboa (IST/UTL), through the Carnegie Mellon Portugal Program, funded by the Fundação para a Ciência e a Tecnologia. In the fall 2010 she went to Carnegie Mellon University. In looking back, five months behind, she was able to go by several challenges: “I wrote two papers, passed my qualifier exam, was teacher assistant in a Robotics Lab for undergraduates, and finished all the courses,” said Susana Brandão explaining that “the last four months were a marathon, but now I can be a hundred percent devoted to my research and doctoral thesis.”
During this time at IST/UTL and at Carnegie Mellon, Susana Brandão made some findings; she discovered an algorithm which combines offline with real-time images in a very effective way. She used the robot soccer team, also from Manuela Veloso research team, and made some tests using a regression learning approach. In fact, she described her results in the paper “Detection of Rotational-Invariant Objects through Regression,” wrote with her two co-advisors, João Paulo Costeira and Manuela Veloso, and published in the Proceedings of the 5 th Workshop on Humanoid Soccer Robots @ Humanoids 2010. In their paper, they explained the regression learning approach which consists “in two main phases: (i) off-line training, where the objects are automatically labeled off-line by existing techniques, resulting in learned object models through regression, and (ii) online detection, where a given image is efficiently processed in real-time with respect to the learned models.” The authors showed that “in robot soccer, it is possible to leverage past experience to create simple and adequate models of objects without the need of computationally expensive algorithms nor explicit modeling of objects.” They also found that by accumulating past images and using the current state of the art algorithms to provide ground truth, the robots gained access to an unlimited number of labeled data which can be used for training the coefficients of a regression. The resulting algorithm is faster than the one used for training but without affecting precision considerably. Furthermore, the algorithm is capable of identifying its own error, which allows for online validation of its results.
Therefore, Susana Brandão believes that it will be possible to adapt this algorithm and to make it work on the CoBot robot. The research team expects that Cobot will be a fully autonomous robot that will do multiple tasks in our everyday life. It has a multidirectional base which allows it to roll forward, backward and sideways and a platform that holds a Microsoft Kinect sensor and a tablet PC that runs software.
CoBot2 can be operated remotely or locally through the same web-based interface. Susana Brandão is working specifically with the Microsoft Kinect, which gives CoBot the opportunity to distinguish image depth and light identity. Researcher in computer vision, Susana Brandão will join these characteristics to the research field in object recognition. The goal is that the CoBot can see besides pixels, i.e., Susana Brandão expects the robot to identify the entire environment.
From July 5 to 11, 2011, Susana Brandão made a poster presentation at the RoboCup 2011 about “Fast Object Detection by Regression in Robot Soccer”. The paper was written by this doctoral student with her two advisors João Paulo Costeira and Manuela Veloso.