Industrial Experience

Please note that the intellectual content and links presented on this page are entirely owned and managed by The MathWorks Inc.

Deep Learning on Embedded Devices via Code Generation

Starting R2020a, MATLAB has automated the process of deploying a generated code that exploits the strength of Deep Learning Toolbox onto embedded devices like Raspberry Pi. This helps integrate target-specific libraries (like ARM Compute Library) with many of the Deep Learning APIs in MATLAB. I was responsible for creating an example that illustrates the strength of this new feature that can achieve this entire task of code generation and deployment via a simple command 'deploy'.

The example features a camera mounted on a servo motor that has two independent functionalities:

Determine the age of people in the camera frame in real-time.
Detect and track a specific object: Move the camera such that an object of interest remains at the center of the capture frame vertically and horizontally.

It is worth noting that the computation is performed solely on the embedded target. MATLAB is used to merely create a generated code and deploy the same to the remote Raspberry Pi device. The camera and servo motor are attached to and controlled by the Raspberry Pi itself. Contents from this illustration have been made part of various MATLAB examples, the links of which are provided below:

Advanced example and content creation for modeling and simulating vehicle dynamics in a virtual 3D environment

I was responsible in creating an example that simulates a vehicle driving around an oval track that is specified by waypoints. Only a small subset of the waypoints are loaded and a MATLAB function determines the speed and heading direction between waypoints. Further, illustrations to create one's own track using RoadRunner and Unreal engine are also given. Necessary links for the example and its corresponding tools are given below: