Plan-Based Robot Control

Autonomous mobile robots with incomplete control and incomplete knowledge of their environment profit greatly from knowledge-level representation and reasoning for their rational and goal-oriented action: active sensing, sensor data disambiguation, performance optimization, human-robot co-operation, explanation – these and more aspects of robust and beneficial robot performance are impossible without applying some symbolic knowledge. However, intertwining in closed-loop and in real-time the faculties of knowledge-level reasoning, robot control, sensing and sensor data interpretation is a fundamental challenge in AI. Within this spectrum, research in the DFKI research department Plan-Based Robot Control (PBR) focuses on hybrid (task-level, spatio-temporal, hierarchical, probabilistic) planning, and we are developing control architectures to amalgamate it with multi-modal active sensing, learning-based methods for sensor data interpretation, object anchoring, multi-modal semantic mapping, and closed-loop plan execution and monitoring.

Applications of our basic research in several domains range from developing long-term autonomous robots to integrating the respective robotics technology for assisting humans in current automated processes. The most prominent application domain is agriculture: as a data-intensive as well as knowledge-intensive, highly digitalized area featuring uncertainty and dynamics, it facilitates AI technology to be profitably applied. The Osnabrück region offers an eco-system of innovative medium-sized ag-tech companies and research partners in agri-food, in which our respective research transfer is rooted.