WP2: Robotics and computation methods for production

This work package consists of collaborative research in robotics and advanced industrial production systems. It aims at fundamental advances in the underpinning theories and methods, including discrete optimization, machine learning, decision-making and verification, as well as an effective transfer of results into industrial practice.

WP Leader: Robert Babuška

Research Areas (RA) and Research Objectives (RO)

Lead: Libor Přeučil
RA6 focuses on vision-based navigation for weakly-controlled environments without a dedicated navigation infrastructure. The research will lead to solutions addressing robustness, self-recovery from runtime failures and the ability to handle cases with high uncertainty, variations, and human presence.

  • Robot workspace modelling, robot under uncertainty (Karel Košnar)
  • Perception-based navigation using embedded workspace features (Libor Přeučil)
  • Long-term autonomy, fault detection and recovery (Miroslav Kulich)

Lead: Robert Babuška
RA7 aims at making robots valuable work companions of humans. Current collaborative robots are not flexible, easily reusable or efficient. A modular architecture and knowledge base will be designed to overcome these problems. Novel approaches will be developed to represent demonstrated skills and tasks, and to schedule tasks between robots and humans, including different modes of robot autonomy. The system will also feature modules for interactive perception and multimodal human-machine communication.

Lead: Martin Saska
RA8 focuses on multi-robot autonomy in cooperative industrial production. Cooperative aerial robots (UAVs) can significantly improve future industrial production, e.g., by delivering components inside and outside industrial facilities. Currently, the deployment of UAV teams is limited by the quality of localization and mapping, flight speeds, and the efficiency of distributing tasks among a team of robots. Therefore, the focus will be on developing novel multi-robot mapping and localization techniques, motion planning for UAV agile flight in unknown dynamic environments, and on high-level mission planning for efficient deployment of multi-robot teams.

  • Topological multi-modal mapping and cooperative localization (Martin Saska)
  • Trajectory and high-level mission planning for agile multi-robot flight (Vojtěch Vonásek)

Lead: Tomáš Svoboda
RA9 researches machine learning to make the industrial deployment of robots more flexible. It will focus on weakly-supervised and self-supervised learning methods that respond to the enormous demand for human data annotation. Inspired by biological systems wherein intelligence is tightly connected with an organism’s body, concurrent and distributed reactive control will be researched in combination with whole robot body sensing. The new methods will make it easier for the system to adapt to new working environments, new sensors and new hardware.

Lead: Jan Fajgl
RA10 aims at higher efficiency and productivity in factory logistics and agriculture, using non-myopic planning and self-improving systems. The focus is on combinatorial sequencing and continuous optimization, augmented by the robot’s motion constraints. RA10 aims at quality guarantees with practical applicability in real-life deployments and method generalization to dynamic problems wherein the system’s performance can benefit from understanding long-term dynamics and online decision-making.

Lead: Zdeněk Hanzálek
RA11 focuses on high-performance algorithms using graph theory, (meta)heuristics, mathematical programming, constraint programming, automated planning and machine learning. Attention will be paid to the novel extensions of production scheduling problems, bin packing, energy awareness, industrial communication scheduling and long-term autonomy decision-making. Both model-based and data-driven approaches will be considered in dealing with practical issues such as supply-chain disruption, personnel unavailability and parameter uncertainty.

Lead: Mikoláš Janota
RA12 will advance formal methods to enable scalable analysis and improvement of the software used in robotics and production in general. The scalability challenge will be tackled from the angle of static code analysis, automated reasoning, and theory. RA12 will focus on the development of novel approaches to symbolic execution, and code optimization supported by reasoning tools that automatically adapt and improve based on previous experience. Specific industrial problems will be tackled theoretically, anchored in the field of parameterized complexity.

Lead: Vladimír Mařík
RA13 develop methods for modelling, designing, and controlling manufacturing systems that allow a flexible response to changing production requirements through easy reconfiguration. RA13 will investigate multi-agent modelling to capture the behaviour of complex manufacturing systems and knowledge engineering methods, working towards fulfilling the vision of plug-and-produce. Transferable machine learning methods will be applied to reduce the training data requirements for manufacturing quality management systems.