Human Friendly Robotics: SaphariShots created by: Karol Hausman, Karsten Knese, Ross Kidson, Sebastian Nagel supervised by: Sami Haddadin, Sven Parusel, Kai Krieger, etc. 20.08.2012 1 Agenda Introduction Communication Architecture User Interaction Face Recognition Cup Finding Manipulation Conclusion 20.08.2012 2 Introduction Goal: Automated serving of medicine To the correct patient Flexible Intuitive user interface Error handling 20.08.2012 3 Introduction 20.08.2012 4 Communication Architecture 20.08.2012 5 State Machine 20.08.2012 6 User Interaction: Sound Sound feedback has big influence on human-computer interaction Sensitive for warnings Feeling more comfortable Complete control of the robot based on speech makes the robot more human-like http://www2.research.att.com/~ttsweb/tts/demo.php 20.08.2012 7 User Interaction: Projector Visual communication between Human and Robot Printing instructions on the table Receiving action through computer vision 20.08.2012 8 User Interaction: Projector What the user sees Instruction what to do Feedback of what the PUT YOUR HAND INSIDE THE SHOWN PICTURE (APPLICATION STARTED) user was doing 20.08.2012 9 User Interaction: Projector What the computer sees Bouding box around ROI Grayscale difference above fix threshold triggers application Could be improved by the help of 3D info based on kinect 20.08.2012 10 User Interaction: Projector Implemented as little state machine Button pressed Welcome Screen 20.08.2012 Bubbles Projector State variable Personal screen 11 Face Detection Haar Cascade Classifier Simple rectangular features - Haar features Cascaded classifier to combine many features efficiently 20.08.2012 12 Face Recognition Eigenfaces / Principal Component Analysis (PCA) Reduce dimensionality of data Trained data as combinations of Eigenfaces 20.08.2012 13 Face Recognition Online training Allows quick and easy training Limited number of people for robust demo 20.08.2012 14 Face Recognition Integration of OpenCV with ROS Subscriber callback for Kinect data Challenges: Memory integrity, synchronisation, bandwidth Integration with Saphari System Actionlib Server to process face recognition request Same framework also used in cup finding 20.08.2012 15 Cup Finding Use Kinect → Point Cloud Library RANSAC Plane Detection Removing far away points Removing the plane Euclidian Clustering Centroid estimation Finding the closest cluster Transformation to the world frame 20.08.2012 16 Cup Finding RANdom SAmple Concensus 20.08.2012 17 Cup Finding RANSAC Plane Detection 20.08.2012 18 Cup Finding Euclidian Clustering 20.08.2012 19 Cup Finding Transformation using tf package from ROS 20.08.2012 20 Manipulation Parameterized 'Serve' Bubble From SafeInit to SafeInit, CollisionStrategy 3 Sub-Tasks Pickup, Place – implemented with Grasp Pour – relative Joint movement of wrist each with approach positions Traversal in between (low stiffness, Elbowfield) 20.08.2012 21 Conclusion Goal: Automated serving of medicine ✔ To the correct patient ✔ Flexible ✔ Intuitive user interface ✔ Error handling ✔ Robust prototype application ✔ Integration of all available devices 20.08.2012 22 Conclusion Lessons learnt Integration is non trivial Human friendly involves more than robot safety Good user interface is important 20.08.2012 23 Discussion Are there any questions? 20.08.2012 24
© Copyright 2024 Paperzz
