Publication

Model-based Fall Detection and Fall Prevention for the NAO Robot

Thomas Münder, Thomas Röfer

In: RoboCup 2017: Robot World Cup XXI. RoboCup International Symposium (RoboCup) July 27-31 Nagoya Japan Lecture Notes in Artificial Intelligence (LNAI) Springer 2018.

Abstract

Fall detection and fall prevention are crucial for humanoid robots when operating in natural environments. Early fall detection is important to have sufficient time for making a stabilizing movement. Existing approaches mostly analyze the sensor data to detect an ongoing fall. In this paper, we use a physical model of the robot to detect whether the measured sensor data indicates a fall in the near future. A trajectory for the foot is calculated to compensate the rotational velocity and acceleration of the fall. In an evaluation with the humanoid robot NAO, we demonstrate that falls can be detected significantly earlier than with traditional sensor classification with little false-positive detections during staggering. Falls due to small to medium impacts can be prevented.

Weitere Links

German Research Center for Artificial Intelligence
Deutsches Forschungszentrum für Künstliche Intelligenz