KMR iiwa Autonomous Door Opening

Copyright Fockelmann

For my master thesis I developed a functionality to open doors with the KUKA KMR iiwa robot. I wanted to create a method that works without any prior knowledge about the door’s shape, geometry or dynamic parameters and with different types of typical door handles.

The handles were detected with a camera (Microsoft Kinect) and object detection by a ResNet+Faster RCNN detector. I formulated the door opening problem with the Task Frame formalism. One Unique feature of my method is that it does not need to run with a fast control loop, but can be exectued on ROS with a commercial robot controller that does not allow for deep system access. The method also includes a sequential control scheme, so that it seamlessly integrates with the ROS navigation stack, automatically opening closed doors that lie on the robot’s path.

I published my approach in this ICRA 2019 paper.

Marvin Stüde
Marvin Stüde
Robotics Engineer

My research interests include mobile robotics, task modeling and machine learning