|
Robust Learning from Demonstrations using Multidimensional SAX |
|
Yasser F.O. Mohammad(Assiut University, Egypt), Toyoaki Nishida(Kyoto, Japan) |
| The paper presents a novel approach to learning from demonstration based on the SAX transform after extending it to multidimensional timeseries data.
The proposed approach was shown to be robust to confusing demonstrations and distortions in the data that is inevitable for fluid imitation situations.
The proposed system (called SAXImitate) can be used as a stand alone LfD system or employed in conjunction with statistical modeling methods (e.g. GMM/GMR) and nonlinear dynamics methods (e.g. DMP) in order to increase their robustness to confusing and distorted demonstrations. |
|
|
Generation of Monitoring Signal in Safety Confirmation Type Contact Sensor using Ultrasonic Wave Propagating in Viscoelastic Tube |
|
Seonghee Jeong, Shota Kanno(Osaka Electro-Communication Universty, Japan), Sumito Kashihara(Osaka Electro-Communication University, Japan) |
| A safety confirmation type wide-range detectable contact sensor is proposed. The proposed sensor generates ultrasonic waves from a transmitter installed at the end of a long and slender viscoelastic tube, and detects the waves as safety monitoring signals by a receiver located at
the other end. An estimation equation related to the propagating sound pressure and time of an ultrasonic wave in the silicon tube was derived. It was confirmed that a safety monitoring signal was successively generated in an arbitrary curved tube by setting an appropriate safety threshold. |
|
|
Verification of Usability of Teleoperation System Using Past Image Records by Sharing Information Obtained from External Cameras |
|
Shunsuke Hara, Noritaka Sato, Yoshifumi Morita(Nagoya Institute of Technology, Japan) |
| In recent years, teleoperated robots have been used for various tasks. We have previously proposed a teleoperation system that presents a virtual bird's-eye view to the robot operator. The scene displayed to the operator is generated by superimposing a virtual model of the robot on a past real image, called a “background image”. In this paper, we propose a method of selecting an optimal background image from past real image records by sharing information obtained from external cameras. We verified the usability of the teleoperation system with the proposed method. |
|
|
Audio Bank: A High-Level Acoustic Signal Representation for Audio Event Recognition |
|
Tushar Sandhan, Sukanya Sonowal, Jin Young Choi(Seoul National University, Republic of Korea) |
| Automatic audio event recognition plays a pivotal role in making human robot interaction more closer and has a wide applicability in industrial automation, control and surveillance systems. In this paper, we propose a new computationally efficient framework for audio recognition. Audio Bank, a new high-level representation of audio, is comprised of distinctive audio detectors representing each audio class in frequency-temporal space. Dimensionality of the resulting feature vector is reduced using non-negative matrix factorization preserving its discriminability and rich semantic information. |
|
|
Visual Saliency Based Segmentation of Multiple Objects Using Variable Regions of Interest |
|
Ayaka Yamanashi, Hirokazu Madokoro, Ishioka Yutaka, Kazuhito Sato(Akita Prefectural University/Akita, Japan) |
| This paper presents a segmentation method of multiple object regions based on visual saliency.
Our method comprises three steps.
First, attentional points are detected using saliency maps (SMs).
Subsequently, regions of interest (RoIs) are extracted using scale-invariant feature transform (SIFT).
Finally, foreground regions are extracted as object regions using GrabCut.
Using RoIs as teaching signals, our method achieved automatic segmentation of multiple objects without learning in advance. |
|
|
Estimation of Human Behaviors Based on Human Actions Using an ANN |
|
MAIERDAN MAIMAITIMIN(Okayama University, Japan), Keigo Watanabe(Okayama Univ, Japan), Shoichi Maeyama(Okayama University, Japan) |
| An approach to human behavior recognition is presented in this paper. The system is separated into two parts: human action recognition and object recognition. The estimation result is composed of a simple action ``Pointing" and a virtual assumed object, which has two attributes, one is ``current status" and the other is ``acceptable behavior". Once the human action and object are recognized, then detect whether a vector calculated by human elbow intersected the object. If the vector is intersected, then estimate human behavior by combining the human action and the object attribute. |
|