Project
| # | Title | Team Members | TA | Documents | Sponsor |
|---|---|---|---|---|---|
| 21 | Pressure Detection: Improving Prosthetics Efficacy |
Mickey Zhang Nathan Beauchamp Sihao Chen |
Yuchen He TA | final_paper0.pdf presentation0.pdf proposal0.pdf |
|
| Psyonic is a local startup that is developing an affordable prosthetic hand for people with upper-limb amputations. Currently, they have a completed and working product that uses electromyography (EMG) to enable the patient to operate the hand. However, they face several obstacles in the prosthetic-arm interface. One of their main challenges is that the EMG signal incurs a lot of noise from many sources, such as high impedance of the skin, external shock, shifting of the arm, and more. This results in unintended movement of the prosthetic fingers. After communicating with the Psyonic team, we believe that we can overcome many existing obstacles by replacing the existing EMG model with a model based on pressure sensing. Our project’s proposed solution mainly consists of three components: (1) Circuitry to collect, process, and store the pressure sensor inputs; (2) Machine learning model that classifies different intensity map patterns from pressure sensor reads to a set of hand finger movements; and (3) Microcontroller executing code to interface with the hardware component and run the classification algorithms. (1) The hardware component needs to convert the analog signals from the pressure sensors to digital signals and ready them for processing by the microcontroller. This will involve operational amplification, sampling, quantization, filtering (to reduce noise), and likely storage in registers. Since the PCB needs to successfully integrate into Psyonic’s existing product, it has to satisfy a number of constraints. First, it needs to be small enough to easily fit into the prosthetic casing. It also has to run efficiently in the low-power environment used in their current product. Finally, it will need to be able to read data from a large number of pressure sensors, as many will be required to produce useful and classifiable data. (2) Classification of the pressure sensor data has to at least be on par with the performance of the current EMG model. We need to train a machine learning model that has higher prediction accuracy and lower latency as compared to the current model. (3) The microcontroller will likely be an off-the-shelf component. It will need to have the processing power required to efficiently run the machine learning algorithms, while simultaneously satisfying the requirements identified earlier for the PCB. Overall, we expect this to be a very challenging- but doable- senior design project. Collectively, our group members have prior experience in the areas of machine learning, embedded systems, and sensors/DSP. Although this is not a novel invention, it is an innovation; we hope to help improve Psyonic's product, enabling better living for disadvantaged amputees worldwide. |
|||||