2016 Fiscal Year Research-status Report
Using mobility to localize sound sources in complex environments
| Project/Area Number |
16K16135
|
|---|
| Research Institution | Advanced Telecommunications Research Institute International |
|---|
Principal Investigator |
Even Jani 株式会社国際電気通信基礎技術研究所, 石黒浩特別研究所, 研究員 (40586228)
|
|---|
| Project Period (FY) |
2016-04-01 – 2018-03-31
|
| Keywords | sound localization / sound mapping |
|---|
| Outline of Annual Research Achievements |
In the current state, the robot is able to build a 2D map of the environment, then use that map to navigate in the environment in order to create a 3D map representing the distribution of the sound. The environment is explored using a path planned in advance. Using this 3D map containing the sound information, it is possible to locate the sound sources in the environment.
The 3D map also contains the geometric information necessary to estimate the sound propagation. In particular, using the normals to the surfaces and the received estimated sound direction, the direction of a reflected sound is estimated.
By exploiting the knowledge about the propagation of sound, the position of a close sound source that is not in the field of view can be inferred in some particular geometries.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
By using well established software framework for the robot and the learning part, the research is progressing rather smoothly.
The sound localization algorithm and the map building has been modified to work with the Robot Operating System (ROS). Thus, it is possible to harvest the existing ROS software to implement necessary robotic functions that are not directly related to the core research.
Familiarization with the different Deep Neural Network (DNN) frameworks (theano, tensorflow) is also on-going in order to test the learning approach. Consequently the focus is shifted on the gathering and preparation of the dataset and not the design of the software framework.
|
| Strategy for Future Research Activity |
The focus will be to move form the simple path planning approach by relying on the inference technique to generate the hypothesis for the path planning.
Improving the detection of unseen sound sources at corridor intersections was identified as an interesting application for robot navigation. This turns out to be an important skill for a robot that has to navigate in an indoor environment. Thus, it is planned to develop a specific method to handle intersections with low visibility and incorporate it in the robot navigation system.
Concerning the search algorithm, the detection of lock-up situations where the search for the sound source is inconclusive has to be investigated. The goal is to be able to identify a lock-up and propose a way to escape for the robot.
|
| Causes of Carryover |
There was no expense for travelling to conference this year.
|
| Expenditure Plan for Carryover Budget |
Use for travel expenses.
|
