Explore Projects

Currently Antrea uses a third-party tool called [bincover](https://github.com/confluentinc/bincover) to measure code coverage when running end-to-end (e2e) tests. This tool has been deprecated in favor of the built-in Go coverage profiling tool (https://go.dev/testing/coverage/) starting with Go 1.20, and it is no longer maintained. We would like to remove usage of bincover from the Antrea project and start using the built-in Go tool. Expected Outcome: Complete removal of the bincove dependency. Code coverage can still be measured with the same accuracy when running Antrea e2e tests and the results can still be reported to [Codecov](https://about.codecov.io/).

<p>This project completes the leftover bits of GSoC'17 and adds a new Rack middleware - Server Timing Middleware</p>

<p>The goal of the project is to create some code for International Components for Unicode (ICU) Librariary that compute dates in Tibetan calendar and also to create a new locale and add some metadata to Unicode Common Locale Data Repository(CLDR) to ensure that CLDR can represent the Tibetan calendars and dates.</p> <p>During the course of the project our primary focus was on two widespread types of Tibetan calendar calculations, Phugpa and Tsurphu.</p>

<p>In Robotics, it is crucial to identify humans and efficiently distinguish them. This gives the ability to perform further challenging tasks such as personalized interactions, social navigation and surveillance. Currently, Robocomp only uses face recognition for human identification. The aim of this project is to integrate other modalities, such as rgb image, skeleton-based pose and silhouette of the body. Using these modalities various human identification components, namely gait recognition and person re-identification, will be integrated into Robocomp</p>

<p>Humans and robots coexist together in today's world and to make these interactions more meaningful, one of the capacities that a robot must certainly perform is to understand the people it is interacting with human beings, and sometimes the knowledge of how many people the robot needs to interact with becomes crucial information. The aim of this project is to add upon the existing functionalities of the ‘People Counter’ in the ‘Human Detection Component’ allowing a more enriched interaction between robots and humans. This will allow the robot to make better decisions based on the density of people present in the area.</p>

<p>Swarm robotics is an approach to the coordination of large numbers of robots in order to tackle a given task inspired by the observation of social animals and their behaviour, making individuals tasks to resolve group problems. My proposal to this idea is to implement examples of collective behaviors using swarm robotics strategies in new or old scenarios of the RoboComp RCIS simulator, like collective exploration of the scenario, patterns formation, morphogenesis or another collective activities. Also, there would be a verbal or non-verbal communication between the swarm components in order to send some commands to the crowd.</p>

<p>The aim of this project is to integrate social behaviors into the RoboComp navigation agent. Mobile robots that encounter people on a regular basis must react to them in some way or the other. Traditional robot control algorithms for path planning and obstacle avoidance treat all sensor readings identically, as foreign objects that must be avoided. However, for a mobile robot that operates around people, these traditional algorithms may not follow the social norms. Even simple conventions like move to the right when a person is encountered might not be honored by these algorithms. However, people would want to, or rather tend to perceive robots-particularly assistive robots, as human like. So catering to this need social behavior will be implemented via certain cost functions.</p>

<p>Interaction of a robot with the operator (human) is an important and daunting task. The aim of this project would be to create a clean, modern and modular Graphical User Interface (GUI) for the conversationalAgent component of RoboComp Viriato. The desktop application will be written using the Electron framework. The app would support one or more open source free to use Text to Speech engines like Mimic or MaryTTS. Combining such a TTS engine with a translation engine can enable the robot to converse in any of the given languages with the user. Leveraging various conversational skills provided by the chosen TTS engine like Mycroft’s Mimic (for example, intents, statements, prompts, confirmations, conversational context) can further help make the conversation feel more real and closer to an actual human conversation.</p>

<p>In today’s world, a lot of educated decisions are made using the information collected by robots and are aided by the inferences given by the software agents which make use of the information collected by the other machines and robots. One such crucial field, where the software agents can contribute is “Pedestrian and Vehicle Traffic Monitoring”, where they can help us identify the hotspots and conflict zones like Pedestrian Crossings, Traffic Signals, Traffic jams, etc. The aim of this project will be to achieve this by using detection networks and depth networks or localisation/pose estimation networks or by building upon the existing functionalities like “peopleCounter_SSDCNet” components already present in the robocomp repository. Through this, the information provided by machines like RGB cameras can be exploited maximally to make better decisions based on the density of people and vehicles present in an area.</p>

<p>My proposal aims to create a new component that allows to store the representation of the robot’s world over time, as well as query it. For this, it is necessary to store the graph structure used to represent the knowledge of the world that the robot possesses. In RoboComp, when robots need to perform complex tasks, its behaviour is based on missions. The mission begin from an initial state and through different transitions reaches a final state, that is, the robot completes the mission. The states are represented by a graph. My proposal is to use a JSON structure to store this graph. This allows its storage as a collection of documents in MongoDB. Consequently, each document could represent a state of the robot. Also, this slot is intended to study the viability of Neo4J for this purpose, as far as I am concerned, the best option would be to use the polyglot persistence, that consists in to use the database best suited for the type of data, in our case MongoDB, and in to use the database best suited for the type of queries, in our case Neo4j. Specifically I think that Neo4j Doc Manager could be our best choice.</p>

<p>The main topic can be divided into two parts:</p> <p>+Body and hand detection: I apply the lightweight Openpose model, MediaPipe library, and the optical flow algorithm. The combination can process in real-time and with a lightweight model, so it is suitable for edge devices.</p> <p>+Gesture recognition: There are two main approaches for this problem: image-appearance-based models and pose-based models. These two are different in types of input data. The first approach's input is only images, while the second one's input is extracted poses from photos.</p> <p>+Extension: Currently, the robot can only learn the gesture from the predefined patterns, which appear in the training data. However, the training dataset is not always available for all motions. Therefore, by applying some unsupervised techniques, the robot can recognize some gestures without any supervised dataset.</p> <p>In this project: Four components will be published: BodyHandJointsDetector, ImageBaseGesture-Recognition, PoseBasedGestureRecognition, and UnsupervisedGestureRecognition. I will also code the testing client for each approach.</p>

<p>The ability of a robot to detect, grasp and manipulate objects is one of the key challenges in building an intelligent humanoid robot. For a collaborative robot, dealing with objects is a key part of its job and its ability to manipulate objects starts with the accuracy of grasping. Intelligent control of a robotic arm to manipulate objects is based on detecting the object and understanding its pose. Precise 3D poses of the objects are necessary to achieve a successful grasp on the objects. We can use the available data from the surrounding environment to recognize the objects poses. In our environment, the data provided will be RGBD frames. Recent work in Deep Learning has achieved amazing results in the problem of 3d pose estimation using RGBD data.</p> <p>In this project, we will work to integrate a pose estimation component to RoboComp using some state-of-art work in Deep Learning. The output poses will be visualized in V-REP simulator and used to drive the new Kinova Gen3 arm, in order to precisely grasp and manipulate some objects.</p>

<p>Object detection and recognition are central problems in computer vision literature and essential for a vision based library. Recent advances in Covolutional Neural Networks (CNN)s have made the detection have made the recognition problem tractable for large number of object categories that would have been very expensive with model based classification approaches. In this project I will implement state-of-art object recognition as well as object detection/localization technique for RoboComp library. The implementation will support both real images and rendered images from CAD model. The implementation will be based on CUDA library and allow user to train and test his models. Also, a caffe independent pure RoboComp based implementation for forward pass will be developed for a selected CNN. Another component based on open scene graph will designed that would allow reading of CAD models and interfacing with current RoboComp simulation framework. Selection of rich textured CAD models is an crucial for any detection algorithm. Therefore, the proposed project will accompany a dataset of cleaned and textured CAD models for 5 object categories.</p>

<p>This project aims at providing support for automatic import of events from various event listing sites into Open Event. This import of events is useful in two ways:</p> <pre><code>1.Import events which are not already listed in Open Event. 2.Allow a user to import events – created on other sites – directly into Open Event. </code></pre>

<p>Implementing an EDS module to sync with EteSync, a secure, end-to-end encrypted and FLOSS sync solution that will enable GNOME/Evolution users to easily end-to-end encrypt and sync their contacts, calendars and tasks.</p>

My project’s primary goal is to create unit tests using KUnit for the AMDGPU driver focused on the Display and Compositing Engine (DCE) 11.2, which will be tested on the GPU “RX 580”. The motivation for that comes not only to assert that the APIs work as expected, but also to keep their behavior stable across minor changes in their code, which can allow for great improvement to the code readability and maintainability. There is a great probability that KUnit will have some limitations in regards to testing GPU’s drivers’ functions, so the secondary goal will be to enhance its capabilities. It’s expected that there will be other people working with KUnit on DCN in parallel, so there will probably be some code review to be done as well. I will keep track of my weekly progress on my blog, reporting the challenges I will face and trying to create an introductory material that could help future newcomers.

While SOF has an incredibly robust and comprehensive back-end implementation, the ability for users and developers to access the features quickly and conveniently has not yet been created. This project seeks to create a GUI that allows real-time configuration and control of SOF. The important aspects of this are the creation of generic implementations that work on all current and future supported platforms, as well as keeping ease and simplicity of use as the number one priority.

I aim to upgrade Visdom from a single-user visualization tool into a scalable, multi-user system. Visdom currently operates on a shared global state and lacks features like user context, authentication, workspace isolation. I plan to implement a JWT-based auth system and refactor the architecture to support user-scoped state. Deliverables are multi-user system featuring isolated environments, role based access control, basic collaboration and presentation features, controlled experiment sharing.

<p>The proposal looks forward to implement features such as API auth, calendar export, review system, etc. that could make the project better and more user friendly. The proposal also looks forward to make UI/UX enhancements in the Admin Dashboard to make the process much more intuitive.</p>

<ul> <li>Bassa has been deployed on a single node Kubernetes cluster with good orchestration yet there is a need for a dedicated monitoring and alerting system integrated into this cluster. </li> <li>Bassa has a dedicated API server that serves the purpose of an automated download queue. Writing efficient client libraries would surely be a great use for other developers to easily integrate the automated download queue into their projects.</li> </ul>

<ul> <li>Improve the feature of automatic import of events</li> <li>Implement a system for event analytics</li> <li>Implement machine learning capabilities for Big Data Services</li> <li>Implement machine learning capabilities for Search</li> </ul> <p>Project Goals :</p> <ul> <li>Addition of automatic import of events listed on various other event management websites into Open Event so to make it as a one stop resource of events across the web.</li> <li>Implement the features for event analytics which allows the users to see how is the events distributed among various categories or event types in Open Event application. This may also include to analyse the interest of attendees towards every specific event type and event topic.</li> <li>Addition of machine learning capabilities to have big data services include in the Open Event API Server so to manage the data as our application scales.</li> <li>To implement machine learning techniques in events search to provide the user with more specific and interested events result set after searching. </li> <li>Writing tests for all the features improved and introduced.</li> </ul>

<p>At its current state, the Splitter - which handles the team of peers - needs to be connected using an IP and Port which must be either fixed or previously agreed for both streaming video and peer connection. Which we want to accomplish here is to have a wrapper Web Service that can manage the native splitters and communicate to clients through HTTP request-responses or any other way that can easily get through NATs-, create an Splitter instance and redirect the video to it, assigning a 'friendly URL' to it that can be easily shared. When this URL is visited, it will reply with the IP and port that the peers need in order to connect to the Splitter.</p>

<p>The audience of PySyft largely consists of people who would like to train their model on private data that reside on other devices/locations.</p> <p>Right now one has to manually spin-up Google Cloud Machines, load a PyGrid instance, queue and run training jobs, deposit the results to another long-running instance(Master) and teardown the created instances(Workers).</p> <p>The project aims to implement functionality necessary to automatically spin-up Google cloud machines, load a PyGrid instance, run a training job, and tear down the instance upon completion (depositing the results into another long-running instance). The primary feature will be the ability to run a “hyperparameter sweep”.</p> <p>The sweep function will kill the nodes(Workers) upon completion of training and depositing the results into another long-running instance(Master).</p>