Riga, 30.09.2025. — The second phase of the project “Development of a robotic system for the automated packaging of spatially randomly arranged objects”has successfully concluded, marking significant advancements in automated log processing and computer vision systems. This period focused on the development of test equipment, sensor calibration, and the optimization of object recognition algorithms.
Key results achieved during this period:
-
A. Advanced Optical Setup: Specialized photo and video equipment has been developed, installed, and calibrated for high-precision data acquisition.
-
B. Automated Calibration & Software: A new auto-calibration program and an improved camera calibration algorithm were established. Furthermore, a comprehensive data processing and analysis program was launched, complete with a dedicated user interface.
-
C. Recognition Infrastructure: Primary object recognition algorithms have been developed and integrated into the workflow. An individual log scanning stand was constructed to facilitate detailed data collection.
-
D. Neural Network Modeling: A high-performance neural network model has been obtained, which will serve as the foundation for log recognition and training in future stages.
-
E. Hardware Optimization: The log scanning equipment and software have been accelerated, enabling the collection of larger sample sizes for both real-world training and synthetic data generation.
-
F. Resource Efficiency: A new method has been defined to significantly reduce required computing power by utilizing alternative types of spatial data.
-
G. Robotic Performance: The number of axes for the log removal robot has been reduced. This engineering optimization significantly lowers the unit price while increasing operational speed, making the system more commercially viable for future clients.
-
H. Point Cloud Analysis: A point cloud recognition model has been successfully developed and tested.
-
I. Innovative Packaging: A new packaging method was defined, allowing for a reduction in the maximum required packaging height.
Impact on future research: The results from this phase have fundamentally shifted the approach to log detection. By analyzing significantly smaller data volumes, the system now requires less computing power without sacrificing precision. The integration of synthetically generated data alongside real-world captures will drastically accelerate the data labeling process, allowing R&D specialists to focus on developing even more precise recognition models. Additionally, the upcoming testing of the new bag-feeding solution will further optimize the system by reducing the overall spatial requirements of the packaging process.
Project implementation period: October 1, 2024 – September 30, 2026. Research budget: 638,520 EUR. Recovery Fund: 444,664 EUR.
Next Step: Would you like me to prepare a more technical version of this report specifically for the technical director, focusing on the point cloud and CNN optimization details?
