ِAsetra‑AI

Project Goal

Develop intelligent multi-robot systems capable of autonomous task execution, collaborative decision-making, and adaptive control in dynamic environments. Our aim is to enhance real-time coordination and robustness through AI-driven methods integrated with efficient communication protocols.

Project Overview

This project, conducted in collaboration with the European Union and Heilbronn University, focuses on Connected Multi-Agent Systems (CMAS) using MQTT for IoT integration. We leverage broker-based publish-subscribe communication to synchronize robots, drones, and sensors, enabling data-driven task allocation and environmental awareness.

What We Have Done

• Designed and implemented multi-robot communication architectures based on MQTT.
• Developed SLAM and sensor fusion algorithms for accurate environmental mapping.
• Applied Deep Reinforcement Learning (DQN) to optimize adaptive control and task selection.
• Enhanced Human-Robot Interaction with speech and tactile sensing modules.
• Integrated PID and adaptive control for precision in navigation and manipulation tasks.
• Modeled and mitigated bias in real-time decision-making through AI and data fusion.

Technologies Used

Software & Frameworks: ROS & Gazebo for simulation, Python and MATLAB for algorithm development, MQTT protocol for communication, YOLOv8 for vision-based perception.
Hardware: NVIDIA Jetson series embedded platforms.
Visualization: Real-time data visualization and monitoring tools.

Project Success & Impact

The project achieved robust, low-latency coordination among heterogeneous agents with improved task efficiency and safety. Bias mitigation strategies increased decision accuracy under noisy sensor conditions. The collaboration with the EU and Heilbronn University ensured adherence to cutting-edge standards and practical validation in simulated and real-world environments. These advancements position the system for applications in smart factories, logistics, and autonomous inspection tasks.

Project Goal

To enable early and accurate detection of skin cancer by leveraging state-of-the-art Transfer Learning techniques applied to both dermoscopic and macroscopic images. Our aim was to make skin cancer detection easy, accessible, and cost-free by enabling analysis with any standard camera, helping to avoid expensive equipment and reduce late-stage diagnosis.

Project Overview

This project, conducted in collaboration with the University of Siegen, Germany, developed an AI-powered diagnostic system that analyzes skin lesions using both dermoscopic (specialized skin imaging) and macroscopic (standard photographic) images. The combined approach enhances detection accuracy by capturing complementary features and supports healthcare providers worldwide.

What We Have Done

• Curated a diverse dataset of dermoscopic and macroscopic images representing a wide range of skin lesion types.
• Utilized advanced Transfer Learning on pre-trained convolutional neural networks (CNNs) for precise classification of benign versus malignant lesions.
• Developed robust image preprocessing workflows for normalization, augmentation, and artifact removal to improve model robustness.
• Implemented a two-stage classification method combining dermoscopic and macroscopic image analysis to increase diagnostic confidence.
• Validated model performance with high sensitivity and specificity on multiple benchmark datasets.
• Designed a user-friendly, camera-agnostic interface to enable free, fast, and easy skin cancer screening using any standard camera device.

Technologies Used

AI Frameworks: TensorFlow, PyTorch for Transfer Learning and CNN modeling.
Pre-trained Models: EfficientNet, ResNet, MobileNet for feature extraction.
Image Processing: OpenCV and scikit-image for image enhancement.
Deployment: Flask-based web platform enabling easy access from any device with a camera.
Collaboration: Partnership with University of Siegen, Germany, ensuring clinical validation and practical relevance.

Project Success & Impact

The system successfully provided reliable early-stage skin cancer detection with accuracy surpassing many traditional diagnostic approaches. By enabling detection using any standard camera, the project breaks barriers of cost and accessibility, supporting early intervention and reducing healthcare burdens. This innovation promises to democratize skin cancer screening globally, especially in underserved regions, through integration with telemedicine and mobile health applications.

AI-Powered Laser Cleaning System

The AILaserCleaner project, developed in collaboration with academic and industrial partners, integrates advanced AI and mechatronics for automated surface restoration. It utilizes a YOLOv8-based detection pipeline trained on corrosion, rust, paint, and biological contaminant datasets.

The cleaning decision-making model is built on Deep Reinforcement Learning (DRL) to optimize laser parameters in real-time based on surface type, detected residue thickness, and surface temperature. A closed-loop feedback system reads data from thermal and optical sensors to dynamically control laser pulse width, frequency, and beam displacement with sub-millimeter precision.

The system was deployed using an NVIDIA Jetson Xavier NX with TensorRT optimization and performs real-time inference at 30 FPS. With custom-designed mechatronic control boards and smart safety interlocks, it ensures operator safety while achieving over 90% cleaning uniformity across varied industrial surfaces including stainless steel, copper, and stone.

Major benefits include a 45% reduction in cleaning costs, complete elimination of chemical use, remote operability via a web interface, and scalable deployment in automotive, aerospace, and cultural heritage restoration sectors.

AI-Smart Welding Machine

The AI Smart Welder project leverages real-time adaptive control and deep learning to enhance welding precision, reduce material waste, and ensure structural integrity. Developed with ROS, MATLAB Simulink, and OpenCV, it uses high-frequency sensor inputs including force, arc intensity, and weld pool imaging.

Using a dual neural network architecture one for surface type classification and one for defect prediction the system identifies joint gaps, angle deviations, and surface contamination. It adapts welding parameters like current, wire speed, and pulse rate during live operations through PID-based controllers supervised by a policy gradient learning agent.

Trained on over 50,000 welding samples, the system achieves a 98.5% weld integrity rating on aluminum, carbon steel, and stainless alloys. Custom anomaly detection modules flag inconsistencies, while integrated CNC interfaces and edge AI computation allow decentralized processing on ARM Cortex boards with GPU acceleration.

This project reduced rework time by 60%, enhanced operator safety, and allowed seamless integration with industrial automation lines. It represents a key step forward in autonomous manufacturing powered by AI.

The company has developed advanced Machine-to-Machine (M2M) communication frameworks tailored for Multi-Robot Systems (MRS). These frameworks address critical challenges in real-time data sharing, synchronization, and decision-making among heterogeneous autonomous agents connected via broker-based messaging protocols like MQTT.

Technical Highlights and Innovations:

• Deployment of a robust MQTT-based communication layer to enable low-latency, reliable message exchange between robots, drones, and embedded sensors.
• Comprehensive modeling and experimental validation of communication latency factors, including message payload size, network congestion, and broker routing, to optimize system responsiveness.
• Integration of AI-driven anomaly detection algorithms (leveraging LSTM and CNN architectures) to monitor M2M data streams in real time, proactively identifying communication faults and preventing task misallocation.
• Development of synchronization protocols to minimize task ordering bias, ensuring fair and efficient task allocation based on up-to-date sensor data and system states.
• Utilization of edge computing for local data preprocessing, reducing communication overhead and enhancing scalability and fault tolerance across robotic fleets.

Achievements and Impact:

• Reduced communication latency by up to 35%, significantly improving coordination speed and accuracy among autonomous agents.
• Improved anomaly detection precision above 95%, increasing system robustness against network disruptions and data corruption.
• Established a scalable and modular M2M communication framework adaptable to diverse robotic platforms and dynamic environments.
• Demonstrated enhanced decision-making efficiency and task assignment performance in both simulations and real-world robotic deployments.

This foundational work provides a reliable and intelligent communication backbone that is essential for autonomous cooperation, dynamic obstacle avoidance, and real-time goal achievement within multi-robot ecosystems.