In laboratory environments, small objects often fall onto the floor and need to be collected and separated rather than discarded collectively. This thesis presents a mobile manipulation system based on the MIRTE Master platform that combines navigation, perception, and manipulation to detect and sort objects in a laboratory setting. The system integrates a 4-DOF robotic arm, a two-bin sorting mechanism, an RGB-D camera, MoveIt 2, and Nav2 with SLAM. Different coverage planners and perception approaches were evaluated in practice. The morphology-based skeleton planner performed well in the tested laboratory-like environments but worse in larger open spaces, whereas the spanning-tree planner performed well in larger open spaces but struggled more in tighter environments. The perception pipeline was able to localize and classify representative objects, and the manipulator modifications improved grasp reliability, although the system remained sensitive to lighting conditions, hardware constraints, and synchronization issues. Overall, the work shows that the MIRTE Master is a feasible platform for autonomous laboratory cleaning, while also identifying several directions for further development.

Introduction¶

In laboratory environments, maintaining a good level of cleanliness is often challenging. While autonomous floor-cleaning robots are common in domestic settings, laboratory floors often contain small graspable items rather than dust or hair. A laboratory-cleaning robot therefore needs to detect, pick up, and sort such objects rather than simply sweep debris. Because not all found objects should be discarded, the system should also be able to separate reusable items, such as electronic components, from other waste. The implementation of such a system also provides useful insights into existing robotic platforms like the Mirte Master robot.

The goal of this project was to develop an autonomous laboratory-cleaning robot based on the MIRTE Master platform. The robot should be able to navigate an indoor environment, detect objects, distinguish between categories of waste, and place them in the appropriate bin. This report outlines the development process and implementation of the resulting system.

From the main goal, the following sub-goals were extracted: adapt and enhance the design of the MIRTE Master platform, implement mapping and navigation pipelines for waste discovery, develop custom waste-detection and localization algorithms, and build a waste-processing subsystem capable of classifying, manipulating, and sorting items reliably.

Related Works¶

Related work on autonomous coverage planning and mobile manipulation provides the context for this project. Coverage-planning methods such as morphology-based skeleton planning and spanning-tree approaches have been proposed to generate efficient paths for inspection and cleaning tasks, but their performance depends strongly on environment geometry and map quality Becoy et al., 2025, Gabriely & Rimon, 2001. Recent surveys of mobile-robot path planning and ROS 2 navigation also highlight the practical trade-offs between planner choice, map representation, and computational resource use in real systems Sánchez-Ibáñez et al., 2021, Macenski et al., 2023.

In parallel, research on mobile manipulation has shown that robust object perception and grasp planning are essential for pick-and-place tasks in real indoor environments Stückler et al., 2013. The present work builds on these ideas by combining a coverage planner, a 3D perception pipeline, and a manipulator on the MIRTE Master platform to evaluate how they perform in a laboratory-cleaning scenario. In particular, the study considers object localization and classification as part of the broader task of detecting, collecting, and sorting small items.

Contribution¶

This work serves as a practical application of the MIRTE Master platform for a specific laboratory-cleaning task. The main contribution is to show how the platform can be extended with a waste-sorting mechanism and to evaluate which navigation and perception approaches are most suitable for this purpose. By integrating perception, navigation, and manipulation in a real-world setup, the project provides insight into the platform’s practical strengths and limitations, aswell as its potential for extensibility and modularity.

Roadmap¶

This report is structured as follows. Section 1 (Materials) describes the hardware and software used to build the robot. Section 2 (Theory) explains the background concepts behind the implemented methods. Section 3 (System Overview) presents the system architecture and the main functional modules. Section 4 (Experimental Methods) describes how the implementations were tested. Section 5 (Results) discusses the experimental outcomes. Section 6 (Conclusion) reflects on the project objective and the extent to which it was achieved.