My research is motivated by the goal of creating intelligent systems with an emphasis on learning, which facilitates adaptation to unknown environments. My primary area of research is machine learning and robotics; more specifically, my research focuses on bipedal locomotion and object manipulation and makes contributions to the fields of 3D image processing, reinforcement learning, dynamical systems and optimal control. Some discriptions from my previous projects are presented below and some of their source code can be found on my Github.
Robotic grasping and object recognition were used to serve a meal on the table. More information can be found in the paper with the title of "Object Learning and Grasping Capabilities for Robotic Home Assistants" in RoboCup 2016 book.
Grasping and manipulation capabilities along with autonomous navigation and localization were developed to be used by a wheelchair-mounted robotic arm to serve patients.
A Kinesthetic learning approach is used to grasp an object. By leaning the object visual features the robot is able to grasp familiar objects. More information can be found in the paper with the title of "Learning to Grasp Familiar Objects using Object View Recognition and Template Matching" in Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2016).
The perception, manipulation, and planning capabilities were developed to be used in the European Robotic Challenge (EuRoC). The developed approach proved to be quite robust and efficient, which allowed us to rank first in the Benchmarking phase. The further results will appear in the paper with the title of "Skill-based Anytime Agent Architecture for Logistics and Manipulation Tasks" at ICARSC 2017.
In this video, I use an interactive object view recognition approach and a similarity metric to grasp familiar objects. The object view rcognition can learn and recognize object views in an open-ended learning manner. This demonstrated experimental results reveal the high reliability of the developed approach.
In my internship project for Open Source Robotics Foundation (OSRF), I developed a Gazebo framework to simulate the RoboCup soccer environment. This project was funded by Google Summer of Code 2015 that its description can be found here. As a test bed, a sample soccer agent had been developed so that its main task is to walk in Gazebo environment. The source code of this project can be found in this link.
Walking using variable hip height vs. walking using fixed hip height. The result of an study to generate an energy efficient walking using ZMP based approaches. For more information please see this paper.
Fast walking experiment using a ZMP based approach and variable hip height. The result of an study on how the robot can learn to walk fast. More information can be found in the this paper.
Our team, FCPortugal, participated in the soccer simulation 3D league which was held in international RoboCup competition, Netherlands, June 2013. Some of the soccer skills of the robots are highlighted during the first half of the match against SEU team (From China), It should be mentioned that all of the soccer behaviours and skills from lowest level to highest level have been implemented form the scratch by the FCPortugal team members. For more information, please check this paper.
Soccer skills for our humanoid team: Nao robots play soccer in the quarter final of the Robocup German Open competition 2012
Results form the research done about model-free walking approaches and bridging the gap between simulation and reality. Stochastic optimization had been used to control the humanoid robot biped locomotion. More information can be found in this paper.