The task at course Development of Intelligent systems, was to implement an agent (turtlebot), which will be able to navigate using predefined map, search for persons using face detector, perform simple conversation with them and fetch their objects, that were hidden somewhere in the room. To solve this problem, we used ROS framework and implemented face detector (using OpenCV), speech recognition (using Google Speech API), space navigation (using ROS) and object detection (using PCL). To perform autonomous object retrieval we used Arduino microcontroller, which operated small arm using step motor. Project was completed with colleagues A.L. and D.S.

Report available here.

At course Machine Learning we had to apply some state-of-the-art ML algorithm to a challenging domain. I decided to compare ensemble SVM approach to standard SVM approach on a domain of sales forecasting. You can download the article, that I wrote as a seminar report, here.

The goal of the E-Learning course was to prepare a web game, which will make music theory in primary (and secondary) schools more interesting. With my colleague, J.S., we prepared a quiz targeted at a primary school music theory lessons. Quiz is currently hosted on Heroku and ElephantSQL database provider, running on a free instances, so it may take up to 30 sec to open page.

Webpage available here.

While working for Epilog, we developed an algorithm for an automated high-rack lift with two double tables, also known as a "miniload". System differentiates from a "classic" miniload system by having four container places on the lift instead of one and having double depth bins (two containers per bin) in the rack. This increases number of possible movement combinations by order of magnitude, which required an efficient implementation of movement decision algorithm.

By setting up a so-called mini-load of the high-rack automated warehouse, the company has substantially improved work speed, process accuracy and stock. Due to efficient lift operations and double depth bins, the warehouse capacity has also increased.

Image credit: SCHAEFER website

Warehouse operators and warehouse management systems encounter a broad spectrum of challenges to ensure the shortest possible shipping times to their customers and maintain their position on the market. The picking process is the most time-demanding among all core warehouse processes. Optimisation of picking is crucial for effective warehouse operation, since it represents the majority of warehousing costs.

In my master thesis I analyzed, proposed and implemented optimisations of the picking process in a manual warehouse. I prepared a logical model of the warehouse, which is based on a hierarchical representation of the warehouse locations. Using this model I reduced the number of locations, which have to be considered during the computation of the shortest picking path in the implemented algorithms. I developed a module for picking path optimisation, which selects one of three implemented algorithms based on the problem size and the warehouse type. To further improve the picking process we introduced changes to the semi-automatic forklift operations and developed a graphical web interface for the warehouse picking heatmap.

Thesis available here.