From Information to Action:
A Perspective on Three Decades of Sensor-Based Robotics Research
Two decades ago, I wrote a dissertation on the problem of general-purpose inference of task-related decisions from diverse sensor data. At the time, it seemed that interesting sensor-based robotics was right around the corner. Has it happened, or are we still waiting? On the one hand, there has been remarkable progress on several fronts -- witness the immense advances in sensor-based navigation or computer vision over the past two decades. On the other hand, it would seem there is still no framework for "Generic Perception" that would support interesting and levels of generic robotic autonomy in unstructured environments. In this talk, I will trace some of the achievements in sensing and sensor processing over the past three decades. At the same time, I will suggest why there has been little progress in other areas of sensing and robotics. Along the way, I will discuss some of the relevant related work from the Robotics and Mechatronics Group at the DLR, and how it exemplifies a unique and remarkably consistent approach toward connecting information to action via sensing in a robotics context.