Abstract:
Demonstration-based learning is a powerful and practical technique to develop robot motion control behaviors, which can be further assisted by continuing to learn from experience after demonstration. The first part of this talk will provide a crash course in the area of Robot Learning from Demonstration (LfD). The second part will overview an approach developed in my research, that acquires a motion control policy from teacher demonstration and then adapts the learned policy with corrective feedback. Validation of this approach has been carried out on in two very different robot domains: mobility control for a wheeled robot, and manipulation control for a high degree-of-freedom humanoid. Corrective feedback has been found to improve the performance of a demonstrated behavior, as well as to enable its adaptation different motion control tasks.
Bio:
Brenna Argall is the June and Donald Brewer Junior Professor of Electrical Engineering and Computer Science at Northwestern University (NU). She also holds a Research Scientist position within the Sensory Motor Performance Program at the Rehabilitation Institute of Chicago (RIC), and is an assistant in the Department of Physical Medicine and Rehabilitation (PMR) at NU. Prior to joining Northwestern and RIC, she was a postdoctoral fellow (2009-2011) in the Learning Algorithms and Systems Laboratory at the École Polytechnique Fédérale de Lausanne (EPFL). Her Ph.D. in Robotics (2009) was received from the Robotics Institute at Carnegie Mellon University, as well as her M.S. in Robotics (2006) and B.S. in Mathematics (2002). Prior to graduate school, she held a Computational Biology position in the Laboratory of Brain and Cognition, at the National Institutes of Health (NIH). Her research interests lay at the intersection of robotics, machine learning and rehabilitation; in particular, on learning robot motion control from demonstration and then enriching behavior development with human feedback, with a focus on robotic devices that provides physical assistance.
IVPL alumni Fang Lu has recently joined the University of Science and Technology((USTC) of China as an associate professor. Wide media attention was brought to the academia achievement she had at such a young age. CV
Born after 1985, her achievements in 3D video analysis and coding, sub-pixel rendering, de-mosaicking, super resolution and compressive sensing led her to the faculty track in USTC, one of the most prestigious universities in China.
During and after her stay in IVPL, she did great work on sub-pixel rendering with Professor Aggelos Katsaggelos. A selected list of journals and conference paper is shown below and access to the papers can be found at IVPL Publications, IVPL
1. Lu Fang, Oscar C. Au, “Subpixel-based Image Down-sampling with Min-Max Directional Error for Stripe Display”, IEEE Journal of Selected Topics in Signal Processing (JSTSP), Vol. 5, No. 2, pp. 240-251, April 2011.
2. Lu Fang, Oscar C. Au, Ketan Tang, Aggelos K. Katsaggelos, “Anti-aliasing Filter Design for Subpixel Downsampling via Frequency Domain Analysis”, IEEE Trans. on Image Processing (TIP), Vol. 21, No. 3, pp. 1391-1405, Mar. 2012.
3. Lu Fang, Oscar C. Au, Yan Chen, Aggelos K. Katsaggelos, Hanli Wang, “Joint Subpixel-based Down-sampling and Demosaicking for Bayer Images: A Fast Frequency Domain Analysis Approach”, IEEE Trans. on Multimedia (TMM), Vol. 14, No. 4, pp. 1359-1369, Aug. 2012.
4. Lu Fang, Oscar C. Au, Aggelos K. Katsaggelos, “Adaptive Joint Demosaicing and Subpixel-based Downsampling for Bayer Image”, nominated as “best paper candidate” (Oral Paper and Top 15% Paper), Proc. of IEEE International Conference on Multimedia and Expo (ICME), July 2011.
5. Lu Fang, Ketan Tang, Oscar C. Au, Aggelos K. Katsaggelos, “Anti-aliasing Filter for Subpixel Down-sampling based on Frequency Analysis”, Proc. of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), May 2011.
Prof. Tsaftaris and Katsaggelos have been working on colorizing a painting by Matisse for almost two years. The final colorization of Matisse's Bathers by a River is part of the exhibit Matisse Radical Invention which is now on display at MoMA (NYC). The story of this interdisciplinary effort between IVPL and The Art Institute of Chicago was covered by major online and print media outlets in the USA and Greece, such as:
Quoting the SPIE press release:
"Aggelos Katsaggelos, Northwestern University, United States, for achievements in image and video processing.
Katsaggelos is a leading expert in digital image restoration, regularized image restoration, Bayesian restoration, multichannel image restoration, and their applications to astronomical images, multichannel images, image sequences, and medical images.
He has also made seminal contributions to key areas in video communications, in particular to source- channel coding and rate-distortion methods for video compression and communication, which has become part of video standards. Recently his innovative work has focused on the direction of audio-visual signal processing, where video information is used to augment speech processing, understanding, and assistance for the hearing-impaired.
Katsaggelos’s work on multimedia signal recovery includes developing the first rigorous techniques for modeling and removing the degradations in compressed images and video. His contributions are significant across a broad range of multimedia problems, for example image restoration and recovery, image and video compression, noise filtering, restoration of video sequences, compressive sensing, and computational photography.
With SPIE, Katsaggelos has been instrumental in many ways including as a perennial member of the steering committee of the Visual Communications and Image Processing conference since 1988. He has served as chair of said conference and co-chair of the Artificial Neural Networks for Image Processing conference. In addition he is a reviewer for the SPIE journal Optical Engineering."
Under the guidance of Professor Aggelos K. Katsaggelos, the IVPL has begun a fruitful collaboration with Professor Karen Livescu of the Toyota Technological Institute at Chicago. Professors Katsaggelos and Livescu are working closely with IVPL student Louis Terry on cutting edge dynamic Bayesian networks (DBNs) for Audio-Visual Speech Recognition.
Professor Livescu joined TTI-C in February of 2008 and is currently an Assistant Professor. She completed her PhD in 2005 at MIT in the Spoken Language Systems group of the Computer Science and Artificial Intelligence Laboratory and was a post-doctoral lecturer in the EECS department of MIT from 2005 through 2007.
New students will have the opportunity to work closely with Professor Livescu through current and future joint ventures.
We are extremely excited to work with Professor Livescu and expect great things!