The Arm Design Contest 2019 brought students from many universities and engineering schools across Taiwan, and since all top three winning projects use STM32 microcontrollers, we wanted to learn more about how they integrated our components. The first edition of the Arm Design Contest took place in 2007, and last year judges received 100 entries, each managed by a handful of students under the supervision of faculty. Participants obtained training in June 2019 to help them use development boards as well as interact with the most common peripherals in the industry, such as UART. There were very few hardware limitations. Students could use any Arm MCU manufacturer and didn’t necessarily have to create a custom PCB.
2019 was highly symbolic as it saw the highest number of projects using machine learning, with a total of 8 teams relying on this new approach. We expect that number to rise drastically over time. Similarly, and for the first time, seven teams created a graphical interface with six of them using our TouchGFX framework and one relying on Embedded Wizard. It seems that the students caught on this new trend that sees the multiplication of GUIs for embedded systems. A panel of independent judges evaluated each project. 35% of the total score is for the depth of technology, 20% is for features and practicality, another 20% is for the completeness of the project, 15% is for its creativity, and 10% is for the final video presentation. Let’s thus look at what made the top 3 projects special.
The project that got 3rd place and that was the most massive-looking was a machine that served badminton birdies with precision to help players train. Users control it by either using a GUI on the machine or through a smartphone app that uses Wi-Fi to connect to the system. The players must manually calibrate the prototype, which can then offer various serves (netplay, drive, lob, smash, or long). The students show that they can achieve consistent results with birdies capable of hitting the same spot with a margin of error of only a few centimeters. The machine is also able to serve birdies with enough power to make it interesting for an average human player.
The prototype uses an STM32F746-DISCO board to power the display, the Wi-Fi module, and act as the main host. Another ST microcontroller, an STM32F0+ is also responsible for managing all the motor drivers that feed a birdie into the launcher and those responsible for shooting it. The system is thus fascinating because it is technically straightforward but feature-rich. It showcases that it’s possible to build impressive and award-winning designs without necessarily using machine learning. This project highlights the significant benefits teams can reap from code optimization and tight hardware constraints.
ASEASAVE is a remarkable device that can perform an electrocardiogram with a small cuff measuring wrist artery vibrations. It then relies on machine learning to examine the sinus rhythm and determine if a patient is suffering from atrial fibrillation or flutter, atrial or ventricular premature contractions, congestive heart failure, as well as hypertrophic, cardiomyopathy, or aortic stenosis. The device must still undergo rigorous clinical trials before doctors can use it in the field, but the first results were truly groundbreaking because they made use of artificial intelligence on an embedded system, and it can make ECGs a lot cheaper and ubiquitous. Labs would no longer need technicians trained in the placement of numerous electrodes or expensive recording devices, for instance.
This project is highly symbolic because it is the only one out of the top 3 to use machine learning. Students used an STM32F7 board to process the information from the sensors measuring wrist vibrations. They also relied on STM32CubeMX to generate the initialization code before building their application, and they developed their graphical user interface using Embedded Wizard. The students even compared the performance of the neural network running on the ST platform against a traditional x86 PC and the results show that despite the vast difference in power consumption, it was possible to obtain a satisfying accuracy, thus proving a key benefit of using AI on an embedded system.
The winning team created a batting tracker thanks to a total of seven boards with two on the main arm, one on the torso, one on the abdomen, two on the leg opposite the arm, and one on the bat. Each board measures a specific movement, and a sensor fusion algorithm that runs on a PC processes the signals to determine the optimal batting form. It takes the art of baseball and turns it into a science. The system heavily relies on wireless networks to transmit the information to the PC since cabling all these boards would be cumbersome and hinder the batter’s movements.
To be able to capture information from so many sensors, the students used an STM32F746 base board onto which they attached a daughter board containing a sensor board and a 2.4 GHz wireless transmitter|. The open aspect of our platform enabled them to put both external components onto a simple custom PCB that they then attached to the main ST board thanks to its Arduino port, thus allowing students to work with components they were familiar with to hasten their developments. The sensors communicate with the STM32 microcontroller through an I2C interface, which vastly simplifies data capture and processing. The ST host then uses an SPI interface to send the data wirelessly to the PC through the 2.4 GHz IC. Despite all this, the boards remained compact at 80 mm x 130 mm x 28 mm, daughterboard included.