The thesis project focuses on developing a dynamic obstacle avoidance system for collabo-rative robots using machine vision technology, integrated with ROS (Robot Operating Sys-tem) and MoveIt framework.

The primary aim is to enhance safety and efficiency in shared workspaces where humans and robots interact closely.

The research addresses several investigative questions, including the implementation of machine vision to improve human-robot collaboration and safety, the system's ability to de-tect and navigate around humans, and its real-time responsiveness in dynamic scenarios. The system utilizes 3D depth vision systems and dynamic path planning algorithms to achieve real-time responsiveness in human-robot interaction scenarios.

The vision system is central to the setup, two Kinect 2.0 cameras are used for environ-ment perception, providing RGB-D data. The cameras are calibrated for intrinsic, extrinsic, and hand-eye parameters to ensure accurate 3D mapping. The captured point clouds are processed through filters such as passthrough, voxel grid, and statistical outlier removal to create a clean and accurate representation of the workspace, which is then converted into an Octomap occupancy map. The Universal Robot UR10e serves as the collaborative robot (cobot). It is integrated with ROS using the Universal Robots ROS Driver, which allows for direct control and com-munication between the robot and the ROS ecosystem via Ethernet.

The system utilizes the RRT-Connect algorithm within the MoveIt framework for real-time path planning. This algorithm enables the robot to dynamically adjust its path in re-sponse to detected obstacles, ensuring collision-free operation.

The entire system is configured on a high-end laptop with ROS Noetic running on Ub-untu 20.04.6 LTS.The testing phase involves evaluating the system's performance in detecting and avoiding both stationary and dynamic obstacles. The response time and accuracy of the obstacle detection and path planning components are critical metrics measured during the tests.Results indicate the system's efficiency in detecting and reacting to humans but highlight challenges in detecting smaller or slim objects due to filtering processes. Response time anal-ysis reveals potential limitations in scenarios requiring quick reactions to unexpected events. Despite challenges, the system demonstrates potential in detecting human body shape and effectively generating collisionfree trajectories, aligning with the project's goal of human detection and avoidance.