Frank Kirchner

Frank Kirchner

Auto-generated profile, based on publicly available information

Frank Kirchner

German Research Center for Artificial Intelligence GmbH (DFKI), 28359 Bremen, Germany

Research Focus & Metrics

Academic Career Length

30Years (1997-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
14 December 2017 in Scientific Reports
Verified
Journal Article
Open access
PDF file
SKSu Kyoung Kim
Su Kyoung Kim
EKElsa Andrea Kirchner
Elsa Andrea Kirchner
ASArne Stefes
Arne Stefes
FKFrank Kirchner
Frank Kirchner
Reinforcement learning (RL) enables robots to learn its optimal behavioral strategy in dynamic environments based on feedback. Explicit human feedback during robot RL is advantageous, since an explicit reward function can be easily adapted. However, it is very demanding and tiresome for a human to continuously and explicitly generate feedback. Therefore, the development of implicit approaches is of high relevance. In this paper, we used an error-related potential (ErrP), an event-related activity in the human electroencephalogram (EEG), as an intrinsically generated implicit feedback (rewards) for RL. Initially we validated our approach with seven subjects in a simulated robot learning scenario. ErrPs were detected online in single trial with a balanced accuracy (bACC) of 91%, which was sufficient to learn to recognize gestures and the correct mapping between human gestures and robot actions in parallel. Finally, we validated our approach in a real robot scenario, in which seven subjects freely chose gestures and the real robot correctly learned the mapping between gestures and actions (ErrP detection (90% bACC)). In this paper, we demonstrated that intrinsically generated EEG-based human feedback in RL can successfully be used to implicitly improve gesture-based robot control during human-robot interaction. We call our approach intrinsic interactive RL.

14 January 2019 in PLOS ONE
Verified
Journal Article
Open access
PDF file
DMDaniel Motta
Daniel Motta
ASAlex Álisson Bandeira Santos
Alex Álisson Bandeira Santos
Ingrid WinklerIngrid Winkler
Ingrid Winkler
Bruna MachadoBruna Machado
Bruna Machado
DPDaniel André Dias Imperial Pereira
Daniel André Dias Imperial Pereira
RBRoberto Badaró
Roberto Badaró
Dengue, chikungunya and Zika are arboviruses transmitted by mosquitos of the genus Aedes and have caused several outbreaks in world over the past ten years. Morphological identification of mosquitos is currently restricted due to the small number of adequately trained professionals. We implemented a computational model based on a convolutional neural network (CNN) to extract features from mosquito images to identify adult mosquitoes from the species Aedes aegypti, Aedes albopictus and Culex quinquefasciatus. To train the CNN to perform automatic morphological classification of mosquitoes, we used a dataset that included 4,056 mosquito images. Three neural networks, including LeNet, AlexNet and GoogleNet, were used. During the validation phase, the accuracy of the mosquito classification was 57.5% using LeNet, 74.7% using AlexNet and 83.9% using GoogleNet. During the testing phase, the best result (76.2%) was obtained using GoogleNet; results of 52.4% and 51.2% were obtained using LeNet and AlexNet, respectively. Significantly, accuracies of 100% and 90% were achieved for the classification of Aedes and Culex, respectively. A classification accuracy of 82% was achieved for Aedes females. Our results provide information that is fundamental for the automatic morphological classification of adult mosquito species in field. The use of CNN's is an important method for autonomous identification and is a valuable and accessible resource for health workers and taxonomists for the identification of some insects that can transmit infectious agents to humans.

1 January 2015 in International Journal of Advanced Robotic Systems
Verified
Journal Article
Open access
MHMathias Hoeckelmann
Mathias Hoeckelmann
IRImre J. Rudas
Imre J. Rudas
PFPaolo Fiorini
Paolo Fiorini
FKFrank Kirchner
Frank Kirchner
Tamás HaideggerTamás Haidegger
Tamás Haidegger
Commercial surgical robots have been in clinical use since the mid-1990s, supporting surgeons in various tasks. In the past decades, many systems emerged as research platforms, and a few entered the global market. This paper summarizes the currently available surgical systems and research directions in the broader field of surgical robotics. The widely deployed teleoperated manipulators aim to enhance human cognitive and physical skills and provide smart tools for surgeons, while image-guided robotics focus on surpassing human limitations by introducing automated targeting and treatment delivery methods. Both concepts are discussed based on prototypes and commercial systems. Through concrete examples the possible future development paths of surgical robots are illustrated. While research efforts are taking different approaches to improve the capacity of such systems, the aim of this survey is to assess their maturity from the commercialization point of view.

1 November 2002 in Robotics and Autonomous Systems
Verified
Journal Article
Closed access
BKBernhard Klaassen
Bernhard Klaassen
RLRalf Linnemann
Ralf Linnemann
DSDirk Spenneberg
Dirk Spenneberg
FKFrank Kirchner
Frank Kirchner

1 October 2008
Verified
Conference Paper
Closed access
MEMarkus Eich
Markus Eich
FGFelix Grimminger
Felix Grimminger
FKFrank Kirchner
Frank Kirchner
For disaster mitigation as well as for urban search and rescue (USAR) missions, it is often necessary to place sensors or cameras into dangerous or inaccessible areas to get a better situation awareness for the rescue personnel, before they enter a possibly dangerous area. Robots are predestined to this task, but the requirements for such mobile systems are demanding. They should be quick and agile and, at the same time, be able to deal with rough terrain and even to climb stairs. The latter is always required if the rescue personnel has to get access to higher floors inside a building. A rugged, waterproof and dust-proof corpus, and, if possible, the ability to swim, are only a few of many requirements for such robots. With those requirements in mind, the hybrid legged-wheeled robot ASGUARD was developed. This robot is able to cope with stairs, very rough terrain, and is able to move fast on flat ground. We will describe a versatile adaptive controller, based only on proprioceptive data. An additional inclination feedback is used to make the controller versatile for flat ground as well as for steep slopes and stairs. An attachable float provided, the robot is able to swim, using the same locomotion approach. By using twenty compliant legs, which are mounted around four individually rotating hip-shafts, we use an abstract model of quadruped locomotion. For the control design, four independent pattern generators are used. In contrast to many other hybrid legged-wheeled robots, we use the direct proprioceptive feedback in order to modify the internal control loop, thus adapting the model of the motion pattern. For difficult terrains, like slopes and stairs, we use a phase-adaptive approach which is using directly the proprioceptive data from the legs.

1 August 2017 in Robotics and Autonomous Systems
Verified
Journal Article
Closed access
JFJose De Gea Fernandez
Jose De Gea Fernandez
DMDennis Mronga
Dennis Mronga
SSStefan Stiene
Stefan Stiene
TKTobias Knobloch
Tobias Knobloch
MWMalte Wirkus
Malte Wirkus
FKFrank Kirchner
Frank Kirchner