@eshaanm

Today was the final day and we got everything done which is such a relief. first i ditched the SVM model because it was overfitting to the data or not seeing the difference between rest and a few grips. the LDA was much simpler and less sensitive. i got a 75% accuracy which isnt great but also not bad considering the data was really messy and LDAs arent supposed to get super high accuracies. I tested it on ymself worked fine then i edited a minute long video and also fixed the github. everything should be ready to ship!

today was one of two lock in days. i had to get some data, train a model, and do some real time classification. i started with just two states: rest or clench of fist. that worked well with an LDA model (100% accuracy) and real time worked well too. next i did a larger set with 6 distinct grips. this took some time to collect and clean up and visualize, and i started trying to train models on it. LDAs had too low accuracy and SVMs were overfitting the data; so while they had a 95% accuracy on paper, real-time classification was horrible. I think i have found the main issue after some serious debugging. the issue is that for some odd reason, the rest baseline drifts over the collection period. i have no idea why but since its drifting the entire LDA is getting thrown off at new data. and since teh SVM data was shuffled it looked fine on paper but real time just didnt work. i need to next adjust for that drift, create a video with the footage i created, update the github, polish the project a ltitle more and hopefully get ts submitted tommorow.

today i worked on collecting some V1 data for my band. first i fitted it on my hand which took some time since one of my hands was in the band. the first trial the data did not save to the SD card and half the sensors were accidentally unplugged. the second trial everything looked fine and the serial monitor showed the sensors collecting promsing data. But, after i analyzed the microSD data it seemed like maybe it didnt save properly or wass the wrong file because it showed no change in the sensor values and was so long (100,000). we will try again tommorow, record a video and then a long lock in halloween morning to get it shipped.

yeah so today just testing more functions. tested the OLED and it was working great. Then tested the IMU. the IMu did nbot work, which was fine for my project. Im asuming it was too small so some connections bridged. I might have broken the OLED with the heat gun since i was trying to take the IMU out, but i ordered a new OELD from amazon and the SD card also works goood. next I have to make one final code, collect some data, train a model on that data, and then do some real time recognition and write up everything for shipping. I have 5 days so tis going to be insanely tight. fingers crossed.

very good day. the band is fully finished. i assembled the new band, assembled the rig, tested all the sensors, got some code working that tested the FSRs and the LEDs. looking good so far. lots of debugging though. next i want to test OLED, SD, IMU, etc. build some code aound that and then finally get some predictions going so i can ship by the 31st

Today was a pretty good day. I started to CAD parts of the rig and a knob for the screw. then I partially assembled part of the new band. I forgot to print the carriages so once i get that done i can finish the band. After the band is done, i will finishe the rig, collect data/callibrate with some code, do some real time predections, and get everything ready to ship.

It has been a long time. i reached out to an FMG engineer and he reccomended that i have a quick release mechanism for the band. the issue is that everything was already on my band and since there was superglue it would not come off. that kindof sucker but i had to take everything off of that including the dowels and pcbs. I luckily had extra screws i could resuse. then i spent time cading a new version of the band that would allow the entire top half to hinge out. next we are going to reassemble the band again, build the rig, collect some data, train, and finally ship.

Today was super nice because all the individual parts came together. i attached all the sensor PCBs onto the rigid band + carriages and then screwed on the main board. then i plugged in all the connectors and glued down the FSRs with some adhesive. looking great. next i do want to manage the wires a little better and make a bettery cover. after that we make the rig and the code and it should be ready to ship.

Today I worked on the rigid band V1. I printed out new carriages and assembled everything with the screws, nuts, carriages, and covers. I used super glue. Next I am going to attach all the PCBs and complete the wiring before moving on to code

Today was a productive day. I got the 5 smaller PCBs soldered and then I was able to add n the force sensitive resistors to all 9 PCBS (I had 10 and then lost one idk) and also added the JST connectors to all 9 PCBs.

today i took one of my small PCBs, soldered on a forse sensitive resistor and a JST connector and tested the sensor and board. it worked perfectly even when being powered by the main control board. The sensor was fairly accurate and sensitive and pressing it fully it never went over 3.3v which was from the op amp i put in which means everything works great. next i have to assemble all 10 of the sensor boards and then integrate everything into the final band/rig.

Today was a productive day. I was working on the PCB hardware. First I soldered on the battery connector and switch and tested the buck-boost converter. it was working which was awesome! mThen i soldered on some headers and hot-aired the FSR connectors. I also soldered on the teensy headers and some other stuff. All that is left on the board is to sodler on the screen which i forgot to do. In the future i will probably just be working on some paperwork so this may be last devlog in some time.

Today was definetaly the biggest day this project has had so far. The PCBs arrived and I hand soldered them using a hot air gun for the first time I had ever done this. First i did the smaller boards. I did not have a stencil for those ones so I just put blobs of sodler paste and then placed the components and used the hot air gun. the hot air gun worked VERY well. Then i moved to the big board which was the scary part. for this one i had a good idea to organize the components though. I took one of my extra boards, put some double-sided tape on it and then placed the components ON there in the right places. then i just moved it to the final board which i applied solder apste on using a stencil. i then reflowed and everything clicking into place was vbery satisfying. I had one bridge on the buck boost but fixed that after and then i just cleaned up. Next i want to finish the rest of the small boards (5 left) and also test the boards to see if they even work.

Today I started working on the code. not the one on the PCB (I finished that already) but the actual model that will predict the gestures. I spent most of the time setting up halftime with Arduino IDE which was annoying since I can’t use lapse for code. I mainly just worked on the tool stack I will need. I am thinking python + usb cable/sd card for data + raw values in a LDA classifier + back to the sd card + teensy runs the trained model ON THE DEVICE. I could also run the model on the computer and then send the predicted value back to the teensy with usb but idk if that is good. anyways next I think I will start on the building of everything while I wait and then move to code while I wait for my PCBs (they are in Hong Kong airport rn) then in October after star dance I will do the experiment and methodology.

Today was pretty good because i worked on the CAD for the sliding adjustable rig. I spent most of the time on the stand for the main FMG band, and then put it in an assembly and tweaked a few alignment issues with the heights and sizes of the rests. Then i printed a portion of the rail and the rest so I can test the tolerances of the dovetail slide. those are printing right now but next I will probably print everything out and assemble the rig and also assembl;e the carriages on the fmg band. After all of this mechanical stuff all i have to do is polish the methodology and the code while i wait for the PCBs to arrive (they will take two weeks minimum so its going to be tight around the end).

it has been some time since my last devlog since school just started and all my PCBs and components are still shipping from china. I thought i would try to finish some of the other work while I wait in order to make it easy and quick once the baords get here. Today i wanted to work on the RIG that will hold the band aw sell as two adjustable rests for the arm. Ofcourse you can use the band without any of this but since this is a research experiment I want to remove as many variables as possible and keeping it in this way locked in one position does that and also makes it easier to test on peple and put it on. Most likelyfinal version will use IMU to correct for limb position and wont need this bulky rig but we will see. Im thinking of keeping the rig on a piece of wood and usinug a dovetail to slide around the FMg rest and arm rests. i modelled two versions. I still need to work on the FMG rest which has to be adjustable up and down but yeah i will try to get that built soon and also finish programming everything and the methodology so i am ready

I added the heat set inserts to the 3d print, soldered the headers on the teensy, ordered both the PCB and it’s components (will take 1 - 2 weeks), and 3d printed the carriage section of the captive leadscrew assembly

today I finished tweaking the PCB then I tried ordering the PCB and LCSC components. I was trying to do this thing where you combine orders to save on shipping, so I had to set the JLC PCB order address to the LCSC warehouse, and then I emailed LCSC and asked to combine my orders since I did not see the option on the LCSC checkout. I will hear back from them soon hopefully so then everything will come so I can star building. also I printed out the 3d printed parts and also got all my other supplied from amazon. I am a bit worried though because one jlcpcb order I cancelled and I do not see the refund on my HCB account

What i worked on today was I mainly was just uploading my reciepts for HCB then I had to go through the schematic and check all the circuits to make sure it was good for printing. I am glad I did because i found multiple mistakes with the SD card, IMU, and buck/boost. I then had to add some more resistors so i had to figure out how to reroute everything. i finished i think but it is still a mess so i can definately work on it for one more day before ordering.