A team from MRS CTU and Fly4Future has successfully demonstrated a drone capable of understanding natural-language instructions, planning its own mission and executing it fully autonomously — without GPS or pilot intervention.
The F4F & CTU SkyGraph team took part in the final of Germany’s SPRIND Fully Autonomous Flight 2.0 competition, competing against seven European technology teams. The challenge was to develop a system that could receive instructions in natural language, understand the task and autonomously determine how to accomplish it. In the final demonstration, the team achieved the best performance in the “find and help an injured person” challenge and again delivered the strongest performance among the finalists.
The developed system combines four cameras, a 3D LiDAR and onboard computing with autonomous localisation and mapping, object and situation recognition, and mission planning based on natural-language instructions. Instead of programming a specific flight path, the user can simply describe the desired outcome. The system then builds an understanding of its surroundings, identifies relevant objects and situations, plans the necessary actions and executes the mission autonomously.
Such capabilities could enable new applications of autonomous aerial robots in areas including search and rescue, firefighting, security, infrastructure inspection and logistics.
The technology is being developed by Fly4Future and researchers from MRS, building on research and expertise developed within the CTU Multi-Robot Systems community. The project demonstrates how fundamental research in autonomous aerial robotics can progress towards technology with practical, real-world applications.
Further details about the system will be presented in upcoming research publications.
Read the full story on the FEL CTU website