MID-PROJECT PLANS
As we worked through our project, a few problems we would need to tackle became clear. The solutions we implemented to are discussed below.
TRANSITION TO TEENSY MICROCONTROLLER!
We are going to be moving from using the Arduino Uno to the Teensy microcontroller. Our reasoning for this is that we need an efficient system with quick computational time in search of having the most responsive system we can. Response time is everything, and the Teensy microcontroller has a 32-bit 180MHz processor compared to the Arduino Uno’s 8-bit 20MHz processor. This difference in processing power will allow us to have quicker processing of data and allow for faster response in our system.
MULTI-SENSOR INPUT!
We were pleasantly surprised by the simplicity of the phototransistor’s signal, but it cannot be used in locations other than the bicep for our recipient since no other upper-arm muscles significantly deform when flexed. Thus, we need to use a second sensor such as an EMG or capacitive EEG sensor to allow for multi-sensor input. This would allow independent control of the fingers instead of having to switch modes to operate fingers, allowing for far easier use. Thus, we will be further researching both sensors to determine which is more feasible, favoring capacitive EEG sensors due to the lower price and potential reliability from our preliminary investigations.
CONTINUED SMOOTHING OF DATA
The moving average filter was applied to the “high motion” data to reduce random noise while retaining important information. So far, this is the only data set that the moving average filter has been applied to. Moving forward, we will be applying the moving average filter to more of our obtained data in hopes to remove random noise for a more distinguished signal.
IDENTIFYING FAILURE MODES
To create the most robust system possible, we will also determine new failure modes for our system that our algorithm may not account for. So far we have tested noise while
Dancing
During high and low levels of motion
Under bright and dim lighting conditions
IDENTIFYING FAILURE MODES
Due to the EMG and capacitive EEG data complexity, we are also going to continue researching and developing the quickest change detection algorithm along with moving averages, frequency analysis, and band-pass filters to isolate resonant frequencies seen only during flexing. These will allow more reliable classification of inputs as flexes or just noise, making the arm feel like a direct extension of the user’s will.
CUMSUM ALGORITHM FOR MICROCONTROLLER
As noted in our progress, we are planning to use an adaptive CUMSUM algorithm based on a collection of a few papers analyzed. This is more false-alarm-safe than moving average, and it still enables us to use a variety of additional digital signal processing techniques.
