جامعة بابل

Patients with severe speech and motor impairments often face difficulty expressing basic needs, especially when they cannot use speech, hands, or conventional input devices. This project presents Amal, an AI-based assistive communication system that enables patients to communicate with caregivers using eye blink detection and sign language recognition. The system uses the mobile camera to detect short and long eye blinks and convert them into dot/dash patterns. These patterns are mapped to caregiver-defined commands, allowing an unlimited set of personalized requests according to the patient’s daily needs.

The Autonomous Rescue and Emergency System (ARES) is an intelligent Unmanned Ground Vehicle (UGV) engineered for search and rescue operations, hazard evaluation, and tactical exploration in disaster environments. Built on an omnidirectional Mecanum-wheeled drivetrain with closed-loop encoders, ARES features a dual-processor configuration combining a Raspberry Pi Pico 2 W for low-level sensor telemetry (GPS, gas, temperature, and ultrasonic distance) and an ESP32-CAM co-processor driving a dual-axis pan-and-tilt servo turret. At the Ground Control Station (GCS), a Flask-based server coordinates real-time telemetry tracking and streams live video processed via edge-accelerated YOLOv8 models for automated fire and survivor detection, utilizing a thermal fusion algorithm for low-visibility operations.

Abstract Modern vehicles rely on the Controller Area Network (CAN) bus for real-time communication between their electronic control units. However, the CAN protocol was designed without authentication or encryption, leaving it exposed to cyber-attacks that can compromise vehicle safety. This project presents CANGuard, an intelligent intrusion detection system that classifies in-vehicle CAN traffic as either normal or as one of three attack types: Denial-of-Service (DoS), Fuzzy, and Impersonation. Instead of depending on heavy deep networks or on raw message timestamps, CANGuard extracts lightweight behavioral features from a sliding window of CAN messages and classifies them using a Gradient Boosting model. The system runs fully offline on standard Windows hardware, responds in about 4.3 milliseconds per decision, and explains every alarm in human-readable terms. Evaluated on unseen test data using a strict temporal split, CANGuard achieved 100% overall accuracy and a macro F1-score of 1.000. The results demonstrate an accurate, fast, and explainable solution suitable for real-world automotive deployment.

وصف المشروع باللغه الانكليزيه This paper is about CyberArena, a kind of training platform for cybersecurity that uses artificial intelligence. The problem is that most training platforms have the old challenges that do not change much so CyberArena tries to make things more realistic.Existing platforms like TryHackMe and HackTheBox have challenges that learners can solve once and then remember, which is not very helpful in the long run. CyberArena is different because it is a website that uses language models and Python to make new cybersecurity challenges whenever you need them. These challenges are in six areas: cryptography fixing code vulnerabilities looking at logs finding vulnerabilities, steganography and exploiting websites. The way it works is that the language models make the challenges and the Python modules handle the parts. This helps to avoid mistakes in the algorithm. CyberArena also has a mode where you can play against someone else in real time. It uses something called Server Sent Events. There is also a recommendation engine that suggests what you should do next a chatbot that helps you and a way to get a certificate that says you are good at cybersecurity. When we tested CyberArena it made 35 challenges in all six areas. Most of them were good to go from the start. It took 5 seconds to make each challenge. This is pretty fast. It means that CyberArena can make a lot of new challenges quickly. CyberArena is a CyberArena platform that can help people learn about cybersecurity in a realistic way with CyberArena being the key, to making it all work. وصف المشروع باللغه العربية تقدم هذه الورقة البحثية منصة CyberArena، وهي منصة تدريب هجينة للأمن السيبراني مدعومة بالذكاء الاصطناعي، تسعى لمعالجة ندرة التدريب الديناميكي، الذي غالبًا ما يكون غير متكرر، بحيث يمكنها محاكاة سيناريوهات الهجمات الواقعية. تعتمد الأدوات الحالية مثل TryHackMe وHackTheBox في الغالب على تحديات ثابتة مصممة يدويًا، حيث يحلها المتعلمون مرة واحدة ويحفظونها بشكل أساسي، مما يبطئ نمو مهاراتهم على المدى الطويل. تقوم فكرة CyberArena على منصة ويب متكاملة مدعومة بنماذج لغوية كبيرة (LLMs) ومحركات بايثون حتمية، تُنشئ تحديات جديدة للأمن السيبراني عند الطلب في ستة مجالات متخصصة: التشفير، وإصلاح ثغرات البرمجيات، وتحليل السجلات، واكتشاف الثغرات، وإخفاء المعلومات، واستغلال الويب. تم تقسيم البنية بطريقة هجينة، حيث يتم توليد المحتوى في جانب النماذج اللغوية الكبيرة، بينما يتم التعامل مع التنفيذ التقني بواسطة وحدات بايثون موثقة. يهدف هذا الفصل إلى تقليل مخاطر التشويش في الأجزاء الخوارزمية. يُقدّم CyberArena أيضًا وضعًا تنافسيًا فوريًا فرديًا، باستخدام أحداث الخادم المُرسلة (SSE)، بالإضافة إلى مُحرك توصيات ذكاء اصطناعي تكيفي، وروبوت دردشة مُرشد بالذكاء الاصطناعي، ونظام شهادات رقمية مُوثّقة. في تجربة توليد داخلية حيث تم إنتاج 35 تحديًا عبر المجالات الستة، حقق النظام نسبة تحقق من المحاولة الأولى بلغت 94.3%، ونسبة تكرار عناوين بلغت 2.9%، بمتوسط وقت توليد 5.17 ثانية لكل تحدٍ.

The increasing demand for intelligent robotic systems capable of operating in hazardous, remote, and inaccessible environments has accelerated the integration of robotics, embedded computing, and artificial intelligence technologies. This project presents the design and implementation of a remotely operated mobile robotic rover that enhances operator safety while providing continuous situational awareness. The proposed platform combines a Zynq FPGA-based processing system with a Raspberry Pi 5 for edge AI processing and advanced sensor integration. The rover incorporates wireless telemetry, GPS positioning, real-time video transmission, and a tracked crawler chassis to ensure reliable mobility over uneven terrain. System control and sensor interfacing are implemented using MicroPython, enabling efficient development and integration of embedded functionalities. To enhance autonomous monitoring capabilities, a Convolutional Neural Network (CNN) model is developed and optimized using TinyML techniques for deployment on the Raspberry Pi 5. The integrated platform, which also includes a live video stream and a six-degree-of-freedom (6DOF) robotic arm, an MQ-2 gas sensor, and a firefighting hose, provides an effective solution for remote surveillance, environmental monitoring, industrial inspection, emergency response, and AI-driven research applications.

This project addresses a significant challenge in the healthcare field, particularly in Intensive Care Units (ICUs), where continuous patient monitoring and early detection of critical conditions are essential, especially for patients with respiratory diseases. The proposed solution integrates artificial intelligence (AI), the National Early Warning Score 2 (NEWS2), deep learning-based chest X-ray analysis, explainable artificial intelligence (XAI), robotic assistance, and automated Telegram notifications to provide real-time clinical decision support and improve patient safety. The expected outcomes include enhanced diagnostic accuracy, faster emergency response, reduced workload for healthcare professionals, and more efficient and reliable monitoring of ICU patients.

This project addresses a significant problem in the digital e-learning domain, specifically the passivity of traditional reading methods and the lack of real-time comprehension feedback for students. The proposed solution is based on a dual-platform system (Web and Flutter Mobile) integrated with advanced Large Language Models (LLMs) such as Gemini 2.5 Flash and Groq/Llama 3.3 70B , with the aim of transforming standard PDFs into interactive, responsive educational resources. The expected outcomes include a substantial improvement in student retention rate, automated personalized feedback, and streamlined lecture management for instructors.

This project addresses a fundamental problem in the medical field: delays in the diagnosis of medical conditions. The proposed solution involves developing an integrated software platform powered by artificial intelligence models trained on specialized medical datasets to aid in the early detection and diagnosis of pathological patterns. Expected outcomes include enabling physicians and healthcare providers to make accurate treatment decisions in record time by providing immediate, automated diagnoses that contribute to improving the quality of healthcare.