A group of German scientists at the DLR Institute of Robotics and mechanics have created an advanced robot and taught it how to sense and catch items in realtime.
Berthold Bäuml, from the German Aerospace Center (DLR) is one of the leaders of the project. The 'Agile Justin' robot is part of research to develop fast perception, catching strategy, body control and dexterity in robots.
For a number of important robotic technologies, it is this kind of dexterity which is necessary to improve the functions a robot is able to perform.
The robot's architecture consists of a range of sensors, computing resources and realtime-monitoring components are all implemented by using DLR's aRD (agile robot development) software concept. The project page states:
Headmounted HD stereocameras track the thrown balls. The rapid catching motion introduces vibrations to the system, which the control can not cancel completely. Therefore, a supporting inertial measurement unit is used for predicting the ball's trajectory.
During the flight the predictions continuously improve and a realtime path planner decides where, when, and in which configuration to kinematically optimal catch the ball. For this, a non- linear optimization problem (including simple collision avoidance) is repeatedly solved on an external computing cluster (32 cores).