RNDr. Petr Štěpán, Ph.D.

Multimodal data like RGB images, depth images, thermal images, Lidar measurements are vital for autonomous systems like aerial robots. The UAV, especially small UAV robots, are limited by sensors and computers weight, so new approaches for data fusion and data preprocessing are necessary. The specialised HW accelerators for Deep NN can be used together with improved data fusion methods.

The relative localization of drones in a swarm is the key to swarm behavior algorithms. The quality and ability to locate each other are one of the important parameters of all swarm drone control algorithms, especially for dynamic systems that allow robots to join and leave the swarm. The aim of this research is to incorporate new sensors based on UWB technology, enabling better mutual detection of robots. Based on the improved system, new algorithms for controlling drones in a swarm will be developed and tested. Research on novel methods and algorithms for mutual localization will proceed in the following directions: (i) Utilize sensors for distance detection based on UWB technology for relative localization. Test the possibility of localizing one sensor relative to multiple sensors by modifying the communication protocol. Enhance the quality of UWB sensor data by integrating it with data from other sensors, such as UVdar, visual cameras, or event cameras. (ii) Another goal is to enable system scalability and adaptability to dynamic changes in swarm composition. The research will focus on designing algorithms and frameworks for real-time data processing that enable agile control and collaboration with the limited computing resources available on board the UAV. The research will focus mainly on the use of GPU and FPGA devices for mutual localization computation and data preprocessing. The research will also focus on enhancing the reliability of mutual localization by verifying local detections between neighbors, which will enable more accurate relative localization in a network of interconnected robots. (iii) The design of algorithms promoting mutual localization of drones with limited communication means, enabling more accurate mutual localization. Communication between drones is limited by distance, and various configurations of robot interconnection topology will be explored to increase the accuracy of localization for all members within the swarm. Theoretical and experimental analyses will evaluate the accuracy, robustness, scalability, and adaptability of the proposed methods in various environments and dynamic scenarios, providing insights and guidelines for enhancing the coordination and performance of large UAV swarms.

Currently, swarms of UAVs, comprising several dozen robots, are becoming increasingly common. Such large swarms are very difficult to control, and methods for effective and user-friendly control are currently being intensively researched. The primary focus of research is on developing methods to examine human interaction for controlling swarm systems in real-time using immersive and multimodal interfaces. The aim is to design new swarm behavior algorithms and methods for involving human operators in these algorithms, enabling operators to effectively control the collective behavior of the swarm and its spatial organization. Research on these topics will include: (i) proposing metaphors for interaction and feedback mechanisms that enable users to control swarm behavior in an intuitive, understandable, and effective manner. Examples of metaphors are the use of gestures to control swarms or the creation of virtual objects in an environment, such as lines, which provide hints to the swarm on how to move through a given part of the environment. The research will investigate how human intent can be translated into swarm dynamics, thereby facilitating effective coordination between humans and swarms. (ii) Current swarm behavior algorithms were not designed for interaction with humans. At this point, the research will focus on modifying these algorithms to enable interaction with humans and designing completely new algorithms, leading to more effective influencing of swarm behavior through interaction with a human operator. The effectiveness of swarm behavior algorithms and their control by human operators will be intensively tested in simulation environments. Additionally, selected algorithms will be tested in real-world environments with actual robots. (iii) Development of control schemes that are adaptable and applicable across different platforms and swarm scenarios. Research will be conducted on the integration of multiple input and feedback modalities, including gestures, spatial inputs, and haptic or visual stimuli. The work will also result in the creation of a framework that will enable the testing of human-influenced swarm behavior in simulations and for real swarms of flying robots.

Design of new algorithms and techniques for onboard localization of autonomous aerial robots. The algorithms have to use only sensors, that are suitable for aerial robots. Another limit comes from limited computation power and storage sources of aerial robots.