Prev-Talk | Next-Talk | All-Talks | Talks-Sorted
Talk-07: Development of a Testbed for On-Board Signal Processing
Mathieu Kemp, Bluefin Robotics
Li Li , Jonathan Odom, Chris Potts, Jeffrey Krolik, Duke University
The common approach to developing signal processing algorithms for hardware is to code algorithms with Matlab and tune them against existing data sets, then convert the code to c/c++ and port it to hardware. The last step is time-consuming, and has to be repeated every time the algorithm is upgraded.
We recently experimented with a method to simplify this procedure, eliminating the need to convert the code to c/c++. The key to the method is to run the Matlab algorithms directly on-board the vehicle, and to use MOOS middleware to provide the interface to hardware.
The method was implemented on a ground robot equipped with a 64-element reconfigurable acoustic array and with LIDAR. It was beta-tested on three separate graduate-level class projects by students having little or no prior experience with MOOS or c/c++. We found that (1) entirely different algorithms - 2D beamforming, variable geometry, 1D beamforming with SLAM - could be integrated in a few days, (2) that the immediacy of running algorithms on live data streams reduced integration time, and (3) that algorithms could be changed/reconfigured in-mission without having to reset the vehicle.
In this presentation we will present this approach and discuss its implementation on our robotic testbed, emphasizing the implementation aspects and the acoustic array/LIDAR MOOS-middleware.
Related Material:
Categories:
- Signal Processing
- Octave/Matlab
- Industry
- Ground Robots