Machine Learning

This project, which was published in NeurIPS workshop 2025, presents a machine learning framework for turning evidence from thousands of biomedical papers into a structured probabilistic model of scientific knowledge. We use an agentic LLM framework to extract relationships between biomedical entities, projects them into statistically meaningful covariance matrices, and clusters them using a custom hierarchical GMM (Gaussian Mixture Model) based on KL-divergence. Results on real biomedical literature and cancer...

This project develops a configurable bin-packing simulation environment in IsaacLab. The system uses a virtual dropper to release rigid cubes into scaled KLT bins, while IsaacLab simulates falling, collision, and settling. The environment supports parallel simulation, configurable cube count and size, vector/image/multimodal observations, reward variants, PPO training integration, camera output, and timing benchmarks. The main focus is to validate a reusable IsaacLab simulation and benchmarking pipeline for future reinforcement-learning and...

Current poker tournament strategies often rely on the Independent Chip Model (ICM), which estimates players’ tournament equity based on chip stacks. However, traditional ICM models provide only a simplified approximation of tournament dynamics and are not directly integrated into practical decision-making agents. At the same time, recent advances in deep reinforcement learning (RL), particularly Deep Q-Networks (DQN), have demonstrated strong performance in sequential decision-making problems but have seen limited application...

This project presents a computer vision–based system for measuring and analyzing patients responses to balance perturbations during rehabilitation. The system uses a GoPro camera mounted on a moving cart to capture patients from a frontal view while walking. Using skeletal keypoint detection, image segmentation, and homography-based transformations, the system identifies recovery steps and calculates their length in real-world units as an indicator of patient stability. The system was evaluated against...

This project presents the design, implementation, and verification of a modular real-time computer vision application for Project Aria smart glasses, with an emphasis on assistive wearable technology for blind and visually impaired users. The developed system connects to Project Aria glasses through the official Aria Client SDK, receives a live video stream, processes the frames on a host computer using an Ultralytics YOLO object detection model, and displays the processed...

LUTHER is a cross-platform mobile application for language learning through audio-only spoken conversation. The app is designed for learners who want to improve their speaking skills in a safe, personal, and non-judgmental environment. Unlike most language-learning applications, which rely mainly on text, quizzes, translations, or gamified drills, LUTHER focuses on the part many learners avoid most: speaking out loud. The motivation behind LUTHER comes from a common gap in language...

Reinforcement learning has shown strong potential for learning control policies in robotics and continuous-control tasks, but its practical use is often limited by the cost of collecting real-environment data. In many realistic settings, collecting target-environment rollouts is slow, expensive, and potentially unsafe. Simulation offers a natural alternative because it can generate large amounts of data cheaply, but simulated dynamics are usually biased and cannot be treated as a perfect replacement...

This report presents our work on the PingPong Hawkeye project, whose goal is to implement a proof of concept for a system capable of estimating the position of a ping-pong ball during gameplay using an array of simple cameras. The cameras used in the project are low-cost devices with characteristics comparable to smartphone cameras in terms of resolution and frame rate. During the project, the system capabilities were demonstrated both...

This project presents the development of a multi-agent pathfinding (MAPF) algorithm as a core computational component within the GOCHESS robotic chessboard, a smart interactive platform developed by Particula. In this system, multiple mobile robots operate beneath the chessboard surface, maneuvering magnetic actuators to relocate chess pieces autonomously. The primary objective of this work was to design and implement an efficient coordination framework that ensures collision-free, time-efficient motion for all agents...

This project tackles the challenge of style transfer between a still image and a specifi c object within a video sequence. The approach leverages the SAM2 model for precise object segmentation in video clips, alongside the AdaAttN model, designed for style transfer between two images. Building upon previous research—specifi cally, “Any-to-Any Style Transfer” which focused on transferring style between an image and an object in another image—this project extends the...

Accurate elevation information is essential for many applications such as navigation and defense systems. However, real world elevation often comes from sensors that provide sparse, non-uniform, or incomplete measurements. Under these conditions, classical methods may fail to produce reliable elevation maps. This project evaluates whether a decoder-only Transformer-based model can outperform classical spatial interpolation, when the known neighborhood is sparse and limited. While the Transformer reduced the average error by...

In recent years, significant progress has been made in the field of soft robotics and robot interaction with deformable objects- flexible materials such as ropes, fabrics, and cables. This area is considered one of the key challenges in robotic control and learning, since the behavior of deformable objects is nonlinear, history-dependent, and influenced by multiple external forces, making it difficult to predict and control precisely. The goal of this project...

In this project We evaluate the SLowcalSGD algorithm on the non-convex MNIST classification task using a simple CNN, comparing it against LocalSGD and MinibatchSGD. We further extend SLowcalSGD by adding a STORM-style double momentum mechanism (μ²) and test it on both convex and non-convex problems. Our results show that SLowcalSGD consistently outperforms MinibatchSGD and LocalSGD. Unfortunately, FL-μ² didn’t show any benefit on SLowcalSGD.

The objective of this project was to develop a real-time facial recognition system running directly on the Crazyflie 2.0 drone using the AI-Deck module. Our goal was to train a neural network for face identification and deploy it onto the GAP8 processor so that inference could run fully on-board during flight. Development Challenges and Approach Throughout the project, we encountered multiple technical issues related to the GAP8 deployment toolchain—specifically NNTool...

This project aims to develop a cost-effective, modular system that processes video streams from a camera and translates them into audio output in real time. Designed to func- tion both as a handheld or wearable device and as a component mounted on a robotic guide dog, the system offers flexibility for different use cases. It consists of a portable processor with a GPU (Jetson Nano) and an Al-enabled camera (OAK-D...

This project focused on training the SPOT quadruped robot in NVIDIA’s Isaac Lab simulation environment, aiming to develop walking capabilities at different speeds and later extend them to goal-directed navigation. The main challenge was to enable the robot to learn stable and flexible locomotion control under conditions that realistically simulate the physical world. To address this, we applied Reinforcement Learning (RL) methods, using the RSL-RL library for policy training and...

This project explores reinforcement learning to teach a quadrupedal robot (Unitree Go1) how to block an incoming football. Inspired by MIT’s Dribblebot, our approach extends the framework from ball manipulation to dynamic goal-blocking behavior. We developed a full sim-to-real pipeline: In simulation (Isaac Gym), we designed custom reward functions encouraging defensive actions like intercepting and positioning between the ball and the goal. In the real world, we integrated a YOLOv7-Tiny...

We implemented Gaussian Mixture Model (GMM) to support personalized medicine through network science. By analyzing gene presence data from sick and healthy patients, the goal is to uncover hidden correlations between genes that may indicate disease. Our model assumes healthy patients exhibit uncorrelated gene distributions, while sick patients show strong correlations. A GMM is applied to distinguish these patterns, with prior knowledge scaling to improve accuracy. Performance is evaluated using...

Our project aims to create a leap hand simulation with three fingers to rotate a cube. Initially designed for a model with four fingers, we are adapting it to be compatible with three fingers. The core objective is to train the model to rotate the cube successfully without failure. This project is fundamentally rooted in reinforcement learning (RL), where we have an agent interacting with an environment. In our setup,...

Efficient traffic management at urban intersections is a critical component of modern smart cities. This project focuses on optimizing traffic light policies at a single junction to enhance traffic flow and reduce congestion. Utilizing the SUMO (Simulation of Urban MObility) simulation environment and Python, we implemented a Deep Q-Network (DQN) algorithm to learn and develop an optimal traffic signal control policy. Through the TRACI (Traffic Control Interface) module, the DQN...

This project aims to develop a portable device to support a new sampling pad designed by the Department of Mechanical Engineering for detecting explosive materials. Traditional sampling methods, such as swab sampling and analysis, work well in controlled environments like airports but are less effective in complex settings like battle zones, where explosive residues are often found in low concentrations. Our device tackles this challenge by heating surfaces to release...

In data-intensive applications, it is advantageous to perform partial processing close to the data and transmit partial results to the central processor instead of the raw data. When the communication medium is noisy, it is necessary to mitigate the degradation in the model’s accuracy. In this project, we address the issue of reduced accuracy in DDNN models due to noise in the communication channel that transmits information from end devices...

In this project, we developed an AI-based policy for managing a 4-way signalized intersection using entropy-driven decision making. By measuring the uncertainty (entropy) in future traffic states, our agent dynamically decides when to switch or hold traffic light phases to reduce congestion and minimize waiting times. Compared to a random control approach, our method significantly improves traffic flow, demonstrating the potential of advanced AI models for adaptive urban traffic management.  

our project focuses on developing a reinforcement learning-based stock trading environment using PyRDDLGym and Proximal Policy Optimization (PPO) agent. The aim is to simulate a financial market where intelligent agents learn optimal trading strategies through interaction with a dynamic environment. This approach leverages Markov Decision Processes (MDPs) and Partially Observable Markov Decision Processes (POMDPs) to model the complexities of real-world stock trading, including price fluctuations, transaction costs, and market volatility....

The Poker Dealer Robotic Arm is a robotic system designed to automate the role of a poker dealer. It handles tasks such as dealing cards and tracking the game state to ensure smooth gameplay. Key Features: • Card Dealing: The robotic arm accurately deals hole cards to players. • Card Recognition: A camera system identifies the cards and their positions using image processing techniques. • Game State Monitoring: The system...

In this project, we developed a simulator that models a solution to the packing problem using the Unity platform. The goal of the simulator is to simulate the process of packing cubes into a container while striving to maximize the container’s volume utilization in a practical and efficient manner. The simulator not only provides a visual representation of the packing process but also serves as a foundation for developing advanced...

Grasping an object using a robotic arm faces many complexities regarding object identification when multiple objects are present, along with the positioning, angle, and gripper key to achieve a good grasp. For this task, several libraries offering computer vision-based models were examined, and the ur5_robotic_grasping library was selected, which uses the GR-ConvNet (Generative Residual Convolutional Neural Network) model. The model provides results that contribute to determining the optimal position and...

Our project which is part of the PINNs research field, was about finding NN architectures that can predict physical phenomenon – in our case a turn of a driving car. During our project we first create some datasets of a turning car, and second we trained MLP architectures to predict it. Our MLP architectures divides into two sections: first, a MLP architecture which should predict the car maneuver without any...

The project aims to develop a navigation platform for autonomous vehicles through an obstacle course, based on a physical model of vehicle motion. The solution integrates numerical methods for solving motion equations or Neural Networks to generate optimal paths in terms of distance and time. The platform considers vehicle dynamics based on Ackerman Steering and allows for real-time navigation solutions. The project employs the PRM (Probabilistic Roadmap) algorithm to generate...

This project aims to investigate the generalization abilities of various reinforcement learning (RL) algorithms across a set of previously untested tasks. Specifically, we will examine the performance of an ExpGen inspired algorithm and compare it with the classical Proximal Policy Optimization (PPO) algorithm. The chosen benchmark environment for this task is Crafter, an open-world survival game with visual inputs that evaluates a wide range of general abilities within a single...

Traffic light control for a single intersection is a well-known problem that affects traffic flow in cities and beyond. There are numerous techniques for managing this problem, ranging from naive policies that rotate through the intersection phases with fixed timings to adaptive policies that sense traffic load on different routes and prioritize accordingly. In this project, we adopt an adaptive algorithm approach that makes decisions based on the intersection’s current...

In many cities worldwide, urban centers serve as transportation bottlenecks, increasing travel times for all road users, regardless of whether their destinations are within or outside the city center. Various strategies have been proposed to mitigate traffic congestion, including the implementation of physical barriers to restrict access, congestion pricing during peak hours, and entrance fees for vehicles entering the city center.   In this study, we propose a solution based...

This project is part of a larger initiative to deploy autonomous ships for oil spill containment. Our focus is on comparing three path planning algorithms by simulating ship movement while considering ship dynamics. We implemented and compared three path planning algorithms: A* (Astar), AO-RRT (AoRRT), and Kinodynamic RRT (kinodynamicRRT). We tested their effectiveness in guiding ships by simulating their movement and tracing the paths while considering ship dynamics. Kinodynamic RRT...

This project focuses on identifying high-activity zones within urban areas to enhance traffic flow and support smarter infrastructure planning. By combining advanced heatmap analysis with K-Means clustering, we’re able to                                                                                ...

This project focuses on making the Segment Anything Model (SAM), better at automatic segmentation and more useful for robotic automation tasks. SAM has been trained on a huge dataset and can handle segmentation well when given manual prompts such as points or boxes. However, it struggles when it comes to fully automatic segmentation without manual prompts. To train the automatic segmentation, the conventional approach creates 1024 points, each generating a mask...

This report presents the development and implementation of a multi-camera multiperson tracking system in an operating room, capable of tracking several people, which aims to enable behavioral research of medical staff members during surgery, in order to improve the understanding of interactions and work processes of doctors and officials during surgical procedures. To this end, we designed and created a software pipeline system that combines advanced technologies in the field...

In our project a pipeline was created to solve the Pick & Place problem for rectangular objects, and a data collection framework was built for further ML and algorithmic purposes. The project relies on the uFactory Lite 6 robotic arm and its software API. An integrated system was built around the arm, including solutions to combine a camera and a vacuum pump installed on the robot. The code framework is...

Given a point robot in a two-dimensional maze environment, the goal is to train an agent that solves the maze consistently and efficiently. At the start of each run, the agent’s initial position and its goal in the maze are initialized randomly. The state and action spaces are continuous, and the reward for each episode is binary—the agent receives a reward of 1 if it reaches the goal, and 0...

This project addresses the critical issue of optimizing traffic flow in urban areas, a challenge that has significant implications for congestion, environmental impact, and urban livability. At the core of our approach is the deployment of a deep reinforcement learning agent tasked with controlling the flow rate of vehicles into a specified urban area. Utilizing the Simulation of Urban Mobility (SUMO) platform as a dynamic environment, we implemented the Proximal...

The positioning of moving objects using trilateration has been in use for many years. There are multiple steps to this process, and in each step, there are multiple methods to choose between to get the best performance. In this project we focused on improving the performance of the existing methods. First, we focused on the first positioning step, trilateration from distance samples received from multiple sensors at a single time....

In our project, we focused on improving the efficiency and management of a four-way traffic intersection using an advanced AI algorithm. We identified the challenge posed by variable and unpredictable traffic patterns from each road, which sometimes lead to congestion and inefficiency in the existing systems. To address this problem, we applied the Monte Carlo Tree Search (MCTS) algorithm to develop a smart traffic control system. Our system dynamically controls...

Introduction The recent surge in popularity of both chess and advancements in AI and robotics technology has led to the innovative integration of these fields in our project. Our goal was to create a fully functional chess game against a robotic arm, providing an interactive experience without the need for computer interaction. Goals The project was built with several objectives in mind, including the accurate detection of a physical chessboard...

To allow for implementation of reinforcement learning algorithms in the game of Starcraft 2, we created an environment that takes programmatic input and uses it to construct an economy and army in the game. With this project, it’s possible to inject a “build order” for the bot to execute as well as dynamically press buttons on a form that would correspond to the bot player in the game taking actions.

In our project, we focused on improving the efficiency and management of a four-way traffic intersection using an advanced AI algorithm. We identified the challenge posed by variable and unpredictable traffic patterns from each road, which sometimes lead to congestion and inefficiency in the existing systems. To address this problem, we applied the Monte Carlo Tree Search (MCTS) algorithm to develop a smart traffic control system. Our system dynamically controls...

The realm of planning and simulation is essential across diverse fields, including applications in drones and aircraft. It involves addressing challenges related to individual agent tasks, coordinating tasks among different agents, and achieving temporal and spatial synchronization. Crucial considerations also extend to the physical limitations of agents and the environment, adding layers of complexity to the planning process. This project specifically focuses on the navigation challenges associated with drone-delivered packages,...

Robotic manipulation demands a seamless integration of perception, planning, and control within dynamic environments. Addressing these challenges necessitates expertise in spatial algebra, kinematics, image recognition, and motion planning. Real-world scenarios add further complexity due to infinite variability. To tackle these issues, reinforcement learning is employed, a tool allowing shifting the focus from solving individual mentioned problems to specifying a goal and a reward function. Reinforcement learning can be categorized into...

Our project sought to improve the manual administration and documentation of anesthesia medications, particularly addressing challenges during emergencies that could result in incomplete medical records. In the initial phase, we conducted controlled injections in a lab setting, specifically focusing on saline solution. Our objectives included automating the detection of injection points, assessing measurement accuracy, and calculating the volume of injected saline.   Transitioning to the second phase in a real-life...

In the context of surgical procedures performed in operating rooms, the anesthesiologist is responsible for manually documenting the administration of each medication administered to the patient. This task becomes particularly critical during urgent operations, where time is of the essence. The dynamic and often chaotic nature of the operating room environment, coupled with patient-specific constraints such as sensitivities and allergies, underscores the importance of accurate and timely medication documentation. To...

The aim of the project is to construct a model for predicting property prices in Israel. This initiative advocates the application of machine learning to develop a predictive model capable of assessing real estate assets with enhanced accuracy compared to conventional methods, ultimately saving time and resources. The predictor takes specific asset descriptors as input, and outputs the current value of the asset. The project spans all phases of the...

– Recommendation systems employ various filtering techniques. The most renowned include content-based, item-based, or user-based approaches. Let’s review some traditional techniques applied to song recommendations using those approaches. – The content-based one identifies songs that share similarities with those already liked by the client, relying on a set of selected features. – In this technique, a predefined set of features is employed to establish an “objective” similarity metric between songs....

Parallel implementations of stochastic gradient descent (SGD) enjoy excellent scalability properties. One of the methods is the synchronous Data-Parallel SGD, a fundamental barrier when scaling the method is the high bandwidth cost of communicating gradient updates between nodes. Consequently, some lossy and biased compression methods have been proposed, by which the gradient updates will be quantized or sparsified. These heuristics are effective in practice but don’t always converge. In this...

In our project, we explored the application of Reinforcement Learning within the context of a smart transportation grid. We learnt fundamental concepts such as Markov Decision Processes (MDP), Reinforcement Learning, and Deep Reinforcement Learning. OpenAI Gym and RDDL were used for the generation of the environment needed for our research process. We wanted our algorithm to enforce cooperation between the agents, making them un-selfish. We hoped that this approach will...

Blackjack is a casino card game in which the player is playing against the dealer. After the cards are dealt, the player can make moves like drawing more cards, split his hand or double his bet. Counting cards that has been drawn gives the player an advantage, as the expected value to win depends on which cards are left in the deck. This project aims to find the optimal strategy...

Cameras traditionally capture a two-dimensional representation of a space, yet many applications, such as robotics, autonomous navigation, and augmented reality, necessitate a three-dimensional comprehension of the environment. Various strategies have emerged to address this need, including specialized hardware like LiDAR, laser, and radar, as well as the use of paired cameras with a known separation for triangulation. However, circumstances may arise, such as budget constraints or the availability of only...

  ALS – Amyotrophic lateral sclerosis, also known as Lou Gehrig’s Disease, is a rare neurological disease that affects motor neurons — nerve cells in the brain and spinal cord that control voluntary muscle movement. Voluntary muscles are those we choose to move to produce movements like chewing, walking, and talking. As the disease progresses, weakness and atrophy spread to other parts of your body. However, the disease doesn’t affect...

ALS (Amyotrophic lateral sclerosis) is an Incurable disease that results in the progressive loss of motor neurons that control voluntary muscles up to complete paralysis. In progressive stages of the disease, patients struggle to communicate with their environment. One attempt at solving this problem is using a brain-computer interface. The goal of our project is to use machine learning to bridge the gap between the resources available to the general...

Task planning contains finding a proper solution to a mission, from beginning to end that contains both discrete decisions (distribution of tasks according to entities, order of events) and continuous decisions (the time which each event is taking place at, planning the trajectory). The task planning problem is a complex problem with undecidable complexity. The ScottyActivity planner’s solutions contain the order of the events and the assignment of the continuous...

The project delves into the realms of reinforcement learning (RL) and machine learning (ML) systems, addressing the challenges these algorithms face in comparison to human problem-solving abilities. While humans instinctively employ intuitive mechanisms like filtering irrelevant data, breaking problems into manageable parts, and adapting problem-solving strategies, traditional RL and ML algorithms struggle to replicate such nuances. To bridge this gap, the project endeavors to enhance RL agents with human-like capabilities....

In multi agent environments, multiple agents perform actions in a joint environment, where the action of each agent affects other agents as well. In this project we developed a custom gym multi agent environment, where the agents are divided into two groups – attackers and defenders with opposite goals. We used two popular state of the art RL algorithms – PPO and IMPALA in order to train the attackers to...

The goal of our project is leveraging a new representation of visual data, Deep Latent Particles (DLP), to train an agent based on Reinforcement Learning algorithm. For implementing and evaluating the project we used Pong, one of Atari games. DLP decomposes the visual input into low-dimensional latent “particles”. Each particle is represented by its spatial location, scale, transparency, and visual features of its surrounding region. Reinforcement learning is a machine...

The gut microbiome refers to the microorganisms that live in the human digestive tract, including bacteria, viruses, and fungi. In medicine, the gut microbiome can be analyzed to diagnose various health conditions. by analyzing the composition of the gut microbiome, doctors can identify potential treatment options, such as probiotics, prebiotics, and fecal microbiota transplantation. A model of the bacteria collected in the microbiome of hundreds of patients as a big...

In recent years, there has been significant progress and interest in the research of generative models. These models learn the distribution of a particular input database and aim to use this distribution to generate new examples. Typically, training a generative model requires a large input database that describes the input space. However, this poses a challenge as constructing such a database is a costly and time-consuming process that requires significant...

PixelNeRF is a neural rendering framework that has shown great success in synthesizing photo-realistic images of complex 3D scenes. Based on the original NeRF, one major addition by PixelNeRF is its ability to generalize its rendering to new and unseen scenes. In this project, we look into how transfer learning might be used to further enhance PixelNeRF’s generalization capabilities. By pre-training pixelNeRF on a different dataset, we attempt to enhance...

When passenger flow is not properly managed in rail systems, it can lead to safety hazards, delays, and overall inefficiencies. By implementing a carefully planned schedule, rail operators can ensure that trains are arriving and departing on time, which can help reduce overcrowding and delays. In order to improve the performance of rail systems, this project presents a control method to regulate disturbed rail system using reinforcement learning.

The project involved using reinforcement learning to create a stock trading algorithm that aimed to maximize the profit of a portfolio. The algorithm mostly received positive rewards due to successful trades, but the learning process was a little bit noisy. Despite this, the algorithm was able to improve the portfolio’s profit over time and gain money.

אחת התכונות הרבות הדרושות לתמיכה בפעילויות של מערכות אוטונומיות היא היכולת לתכנן תנועה. זו מאפשרת לרובוטים לנוע בסביבתם בצורה מאובטחת ולבצע משימות נתונות. לרוע המזל, לולאת הבקרה הכוללת חישה, תכנון ופעולה טרם נסגרה עבור רובוטים בסביבות דינמיות. סיבה אחת כרוכה בזמני הביצוע הארוכים של רכיב תכנון התנועה. פתרון לבעיה זו מוצע באמצעות חישוב מקביל. לפיכך, משימה חשובה היא הקבלה של אלגוריתמים קיימים של תכנון תנועה עבור רובוטים כך שהם מתאימים...

In this project we address the problem of training models and agents to perform various human tasks. The literature dealing with this field offers different approaches that deal with the training problem in diverse ways. All the methods are based on a combination of recognized principles in machine learning such as RL, BC, and original strategies for improving the quality of learning and adapting the algorithms to the nature of...

Reinforcement learning is a powerful tool that lets us solve difficult problems, from playing Go well to have self driving cars. RL algorithms start by defining and environment that simulate the problem we want to solve and some kind of reward that correlate with success. Afterwards the program ”plays” the environment trying to get the maximum reward and using various algorithms it learns from past mistakes. One of the main...

A brokerage account is an investment account that allows you to buy and sell a variety of investments, such as stocks and ETFs, that we will focus on. A stock is a type of investment that represents an ownership share in a company, Bonds – Unlike stocks, bonds don’t give you ownership rights. They represent a loan from the buyer (you) to the issuer of the bond. ETFs or “exchange-traded...

The project research the hypothesis that the GeLU function gives good performance on image classification problems because of the property of not resetting the derivatives of the negative values close to zero in the backward process of the neural network. The reason for testing this feature is that the GeLU function is complicated to implement and expensive in terms of hardware, so the project dealt with the use of this...

Training generative models using per-pixel loss often results in poor generated images. Better perceptual quality can be achieved using feature-based loss functions which use pre-trained models as feature extractors. Unfortunately, pre-trained models usually come with some drawbacks. First, they are often pre-trained on a task that is unrelated to the main task. In addition, they are usually big and computationally heavy. Our project goal was to Investigate the need for...

Our goal was to dive into the GCSL RL algorithm, and expand it based on other concepts in the reinforcement learning field to get a better model, that learns faster, is more stable, and hold some knowledge over how important some states are over others given the goal the agent is trying to reach. We investigated whether Q-values can be introduced into the GCSL algorithm to create a better model,...

The drum kit is a popular instrument in many music genres. However, for new drummers, owning a drum set is expensive both financially and space-wise. Playing drums is typically noisy and bothersome to those around. In our project, we created a software-based drum set that uses only two sticks and a smartphone or a webcam to provide an air-drumming experience. Our solution is implemented by a trained deep network which...

Project Summary: Designed and implemented a Pipe-Line that takes two input images of the user and creates an animatable Avatar in the virtual world. Work Steps: 1. Surveyed existing Directions and NN to find one that fits our needs in avatar and virtual world, Chosen ק– Construct 3D Avatar from human full body image. Investigate network code and improve the following: Improved resolution of output avatar significantly using rendering method...

Muscular dystrophy patients in different stages of the disease have trouble in communicating with their environment. In most muscular dystrophy diseases and ALS in particular, the brain functions normally and so do the patient’s eyes. Today there are systems that allow the patient to communicate using equipment, such as identifying a gaze towards a monitor and creating words and sentences using the identification. These systems cost a lot of money...

Introduction: This project was motivated by the rapid advancement of robot technology. Many robots are becoming self-reliant and able to do tasks autonomously without any supervision. One hard task for a robot to do alone is navigating, and this project was done to see how A star algorithm can solve the problem. AI works in a know environment, meaning the robot will navigate easily if it has the layout of...

In our project we have used crazyflie drone (made by Bitcraze) in order to implement online slim AI network. This is a miniature drone which capable of handling low weights and small sized memory. Therefore the neural network had to be skin. We chose to put our main focus on creating infrastructure to work with NN on the drone (which wasn’t existed fully before) and on creating a basic NN...

Weather is a chaotic physical system that is very hard (computationally and in accuracy) to calculate. Because of its complexity we choose the Dead Sea as a relatively isolated environment as a simple case for this complicated problem. By using the measurements from a meteorological station of the Geological Survey, and the past ECMWF forecasts for the area as our dataset, we tried to use the Transformers (ML) architecture to...

AI – the term that overtook the research and industry world in the last decades, has brought with it some promising forecasts regarding the upcoming future. Predictions of optimization and improvement of the existing technology. And practically, a promise for a fundamental change in the way we live. Image processing via machine learning is one of the main topics in this field. And as such, the motivation to implement it...

Guided by the Vasiliki Plereu et al. paper published in 2002 that incorporated random matrix theory in finding the true correlations of a market of securities, we implement techniques intended to find minimum-risk maximum-return portfolios and sets of correlated or anti-correlated securities. We carry this out with various security datasets obtained from online packages. We go on to test certain limits of Random Matrix Theory (RMT) and propose a real...

Long term planning is a common problem being researched. It appears in a lot of fields such as robot motion planning and navigation which we worked on in this project. In this project the goal was to create an agent that control the motors of a robot joints to walk from an arbitrary point of a maze to an arbitrary goal, in a simulation using SOTA RL algorithms and even...

For years drones have been foretold to change the future of door-to-door delivery, yet they are still far from widespread use. Several factors hold them back from mainstream adoption, such as a small carrying capacity, short operation range, and the complexity of fleet management. Our project aims to tackle some of these challenges by combining control, planning, and deep learning algorithms to allow cooperation between multiple drones for a joint...

In this project we will implement an online algorithm which is based on an existing offline algorithm. The online algorithm receives an optimal program of airplane takeoffs and landings for the workday, and during the day the algorithm will need to handle incidents and changes in real time. The main focus is to handle incidents and changes in real time with minimal deviation from the original program. If the algorithm...

Market trading has become popular among most people these days, though most of them avoid making risky action, because of lack of knowledge and the difficulty to predict its movements. The project goal is to meet the need to ease a portfolio managing, using a system which takes actions in relative short time intervals, with a high profit potential. The system goal is to maximize the profit using US capital...

We propose a method for correcting skinning artefacts caused by Linear Blend Skinning algorithm of triangular mesh object, based on loss that involves the calculation of geodesics distances between all its points to minimize distortion in its shape. One crucial point is the usage of the Heat Method for the calculation of the geodesic distances based on the realization that its pipeline is fully differentiable, in a similar manner which...

In recent reality, people are looking for new ways to connect with each other. Virtual reality can be the new meeting ground for people all around the world. However, capturing a person to view in VR is limited by equipment and capture location. To enable those new applications and usages, we want to achieve a simple setup to enable everyone to participate in this new frontier. In our project, we...

A Rubik’s cube face can be solved for any 3X3 array of six colors. By solving many cubes, it is possible to arrange them into any six-colored image. In this work, we present a python-based cube solver and GUI for generating images from Rubik’s cubes. The GUI provides users with the instructions for all the cubes necessary to create the input image. The application includes three modules. The image processing...

Abstract: The project’s goal was to test the viability of using reinforcement learning for the training of a robot to pick up objects from a surface automatically. We used a WidowX robotic arm in the RepLab configuration that includes a surface on which the robot can practice picking up objects. During the project we discussed three different problems: picking up a single cube from a pre-defined location set given the...

The project’s motivation is to create a robot capable of navigating in an environment with people, detecting the people within range and interact with them. The importance of such project stems from the fact that robots are present in every aspect of our modern life, specifically servicing. Therefore, using “friendly” robots who can recognize people and induce pleasant sensation could be very useful. This project is a continuation of a...

Anisocoria is a condition characterized by an unequal size of the pupils. In some cases, this syndrome is a symptom of another disease that should be treated immediately. Some of the pathologies that may be characterized by anisocoria are – brain tumors, head injury, multiple sclerosis and more. Nowadays, the pupils’ function is examined by a physician. The examination normally includes a rough assessment of the difference between the pupil...

Halite by Two Sigma (“Halite”) is a resource management game where you build and control a small armada of ships. Your algorithms determine their movements to collect halite, a luminous energy source. The game combines both short-term tactics with long-term planning, making it very suited for many Reinforcements Learning \ planning algorithms available today, which is a perfect fit for our goal. Solving the Halite game is not an easy...

Deep reinforcement learning has shown tremendous success over the last decade, but a big whole is yet to be filled until such methods can reliably and safely be deployed into real-life machines. Robust reinforcement learning aims to ensure such safety but is computationally expensive. In this work we focused on implementing an algorithm which is both computationally cheap and has robust properties which can be implemented in the real world....

In recent years there has been significant technological progress in the world of robotics. The need to use robots in operations performed so far by humans has intensified and particularly in tasks that include autonomous navigation of robots, such as bomb disposal or locating missing persons. This project is about the autonomous navigation task. The main goal of the project is to make a Turtlebot2 robot successfully navigate autonomously towards...