Projects
An EMG-Based Virtual Reality Application for Motor Rehabilitation
It is no surprise how VR tech is helping people in many aspects of life. In this work, we focused on bringing more fun and better results against traditional therapy (specifically kinesiotherapy). VR and games sound like a lot of fun, right? But for researches, this kind of application must be followed by experiments and evaluation to substantiate the science behind it.; the reason this project followed a strict methodology and even counted on physical therapist support!
For its achievement, we developed our own circuit board to capture EMG signals (muscles stimulus) and a scary game powered by Unity. After that, we tested with real users, and the results were astonishing!
The work was published in the Symposium of Virtual Reality (SVR), Brazil's main VR/AR conference. Please, check it by clicking here.
Occupancy Grid Map Estimation Based on Visual SLAM and Image Segmentation
Let me introduce you to my baby (masters dissertation haha)! It is regarding feature-based SLAM, which brings efficiency, speed and can offer an accurate location system; on the other hand, the map produced is a sparse representation of the environment, limiting path planning activities and reducing robotic autonomy. The solution presented extends the mapping stage to build an occupancy grid map and a segmented point cloud. For this, it uses the pose estimation from the SLAM system, its sparse map, and an image segmentation technique.
The application needs only a stereo camera, and it can run in conjunction with the SLAM system in real-time while requiring a low memory footprint.
I'm also pleased to share it has been accepted at Latin American Robotics Symposium 2021. Please, check the paper by clicking here, and the dissertation here.
Robotics Simulators and Reality Gap
Simulators are such a fantastic tool, right?! But did you already think how much fidelity exists between simulation and reality? The study behind it is called reality cap. During my journey in Voxar Labs, in partnership with HP, I had the chance to study more about this topic and evaluate two well-known physics engines: ODE and Bullet. First, we built a ramp in which a cube split on it. Second, a real Turtlebot was considered. Coefficient friction in the real environment was calculated for both rubberized ramp and floor. The video illustrates visual results. While real and synthetic environments have similar behavior for the first scene, the simulated environment is much more precise and faster when considering tests with the robot. These simple tests reveal there is space for improvements regarding the virtual and real world.
Autonomous Navigation System
During a discipline called embedded system, two colleagues and I worked on an autonomous driving project that required only a stereo camera for navigation purposes. A ground segmentation technique was embedded in an FPGA board, and based on an image evaluation, the mobile robot decides the route that avoids obstacles. Two batteries powered the entire real-time system, which was developed and integrated from scratch. Later on, this work became the main insight for my master's project. Lots of valuable learnings!
Virtual Reality Game
Individuals in a particular spectrum of autism may face learning challenges (which does not mean a learning disability). Can you imagine if fun alternatives, such as games, could be used to improve this situation? I bet on that! During the Virtual and Augmented Reality course, I led the development of a mobile application aimed to alphabetize autistic children in a more dynamic and fun way. Feedback from experts on the field was used to achieve a pilot version, illustrated in the video.
Biomedical Engineering Challenge
In 2018, the 26th Brazilian Congress of Biomedical Engineering (CBEB) held the first Thematic Challenge in Biomedical Engineering (DETEB). The challenge consisted of controlling a mechanical arm with six joints and 6 degrees of freedom using exclusively electrical stimuli from the muscle (electromyography -EMG signals). The team should conclude a specific activity accurately and in the shortest possible time. Voxar Labs' team assembled the given arm and developed a unique solution. Initially, the use of the MYO armband device was considered and tested by establishing a communication between the device and the V-REP simulator, illustrated in the video. In addition, using machine learning approaches, we expanded the number of interpretations of hand movements. After this first stage, we developed our own low-cost board to capture the EMG signals (more here - page 212). We proudly achieve the 3rd place award in the competition!