Xuan Sang Le is a PhD student who has joined our team mid-february. His work is co-supervized by Ecole des Mines and ENSTA is about speed optimization of Smalltalk robotic software by means of FPGAs. The first step is to develop an application fully in Smalltalk and our PhaROS Robotics framework that will serve as a reference for our metrics. The app we have chosen is a simple tracker robot that follows an object of a particular color. Beside being fun, the result shown by the video below is interesting. As we have initially expected we can’t pretend to be real-time. Still, Pharo is capable of fetching an image through wifi, process it and discover the position of the object 650 milliseconds. And this is with non-optimized code. Figures are likely to be better after optimization.
After less than an hour optimizing the code, we have improved performance with a factor of 5. Now, the full cycle (fetch image, process it and publish the position of the object) takes about 130 milliseconds. So far we did only 2 simple optimizations = perform filtering at the level of the color map + replace costly color conversion at each pixel access with simple access to raw pixel values. Hopefully we can do better.
Great stuff you are doing. Xuan is an IFI student ?
Yes, Xuan Sang LE is a former student from IFI (Hanoi, Vietnam)