You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: website/versioned_docs/version-fils_en/project/2026/mihai.temiac/index.md
+100-5Lines changed: 100 additions & 5 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -48,7 +48,48 @@ As such, I chose this project due to its nature and complexity, as I strongly be
48
48
- First 3D renders, there is plenty of room for improvement, but have been getting more comfortable with Fusion
49
49
50
50
### Week 9
51
-
Working on the documentation.
51
+
- Working on the documentation.
52
+
53
+
### Week 10
54
+
- Got the prints, this is tough. Broke a pin already, the tolerances were not a joke. Will use the universal repair tool, superglue.
55
+
56
+
### Week 11
57
+
- Still working at the build; it feels like a battle, and I am losing.
58
+
59
+
<center>
60
+
61
+
</center>
62
+
63
+
### Week 12
64
+
The assembly has been completed, could say that the superglue was *handy*. You would think that a large hardware store would have all the tools you need, but sometimes you have to improvise (this time, I had to improvise a tad too much).
65
+
66
+
Down below is my fully functional, high precision, low cost, hand drill (could not find a smaller electric drill, and I was not gonna spend the big buck on professional ones, hope I will not be working in construction anytime soon), as the tendon channels needed redrilling.
67
+
68
+
<center>
69
+
70
+
</center>
71
+
72
+
Learnt what an *elephant foot* means in 3D printing; pairing grit paper with my ultra precision tool meant the job was (supposed to) be easy enough. 4 days later, I was still sanding down finger joints, else the servos would snap at the first tug.
73
+
74
+
<center>
75
+
76
+
</center>
77
+
78
+
Also tested the code without the assembly, everything was looking good. Sample sensor, analyze, classify and respond with the correct motors.
79
+
80
+
<center>
81
+
82
+
</center>
83
+
84
+
### Week 13
85
+
Final push: the hand was complete, the assembly looked good, the mechanics ran smooth.
86
+
87
+
<center>
88
+
89
+
</center>
90
+
91
+
For the future students taking this class, please do **demo videos**. You never know when things might go wrong. Installed an I2C module for the servo control, things ran great until they did not. Functional code, got a video of the servos moving, after which we got smoke coming out of the module. Safe to say, I was toast. Document your progress, stuff happens.
92
+
52
93
53
94
### Week 10
54
95
Got the prints, this is tough. Broke a pin already, the tolerances were not a joke. Will use the universal repair tool, superglue.
@@ -64,7 +105,15 @@ The system uses **AD8232** EKG sensors, repurposed for EMG due to the high cost
64
105
65
106
Everything is processed by an **STM32U545 Nucleo** microcontroller. **MG90S** servos will be used in the finger pulley system, as I need enough torque to counteract the force pulling the finger back in its neutral position.
66
107
67
-
Generic string and elastic materials will be used for the tendons, such as fishing line for the flexors and springs for the extensors.
108
+
<center>
109
+
110
+
</center>
111
+
112
+
Two tendons per finger, both via braided fishing line (strongest I could find in store was 24kg, no way the servos snap it). The extensor motion was achieved by tying orthodontic elastics (braces elastics) to a pin mounted in the servo bay, which pulled the finger to its neutral position after the tension was released; almost like a saw tooth.
113
+
114
+
<center>
115
+
116
+
</center>
68
117
69
118
<center>
70
119
@@ -80,6 +129,8 @@ Generic string and elastic materials will be used for the tendons, such as fishi
80
129
81
130
</center>
82
131
132
+
The **PCA9685** has been left out, as the project can run with or without it, via the **PWM** pins on the **STM**. If we were to install it, we would wire up the `SDA` and `SCL` pins, with power coming from the external power supply, (logic 3.3V comes from the STM). The servos are plug and play, provided you do not mess up their orientation.
133
+
83
134
### Bill of Materials
84
135
85
136
<!-- Fill out this table with all the hardware components that you might need.
@@ -98,18 +149,62 @@ The format is
98
149
|[AD8232 EKG sensor](https://www.analog.com/media/en/technical-documentation/data-sheets/ad8232.pdf)| EMG signal acquisition (repurposed EKG module) |[35 RON x 3](https://www.optimusdigital.ro/ro/senzori-altele/1347-modul-senzor-ecg-ad8232.html?srsltid=AfmBOooAX5b3QDFkBnnuSQi5Ejg6U0BX_VEz3xrKOzaeQP8Z8HV6hnwI)|
99
150
|[MG90S Servo Motor](https://www.tinytronics.nl/product_files/000263_Data%20Sheet%20of%20MG90S%20Analog%20Servo%20Motor.pdf)| Finger actuation |[20RON x 3](https://sigmanortec.ro/en/servo-motor-mg90s-metal-gears)|
100
151
| Gel electrodes | Skin-signal interface |[30 RON](https://www.aviafarm.ro/cumpara/set-100-electrozi-adezivi-ecg-ovali-f9089-cu-gel-solid-si-senzor-ag-953?utm_source=portal&utm_medium=web&utm_campaign=google_xml&gad_source=1&gad_campaignid=22826372328&gbraid=0AAAAADQMw-p4DXuVPfSpT3SlDTUKEi_cz&gclid=CjwKCAjwqazPBhALEiwAOuXqdCRNGu_i0Rv1CSS49kxbpT8zRxvhcMcm2q3GkeL3nzO_ce6pKp12-xoCPNoQAvD_BwE)|
| Elastic bands | Extensor tendon system | owned |
102
153
| Fishing line | Flexor tendon system | owned |
103
154
104
155
## Software
156
+
### Signal Acquisition
157
+
158
+
Each **AD8232** connects to the **STM's**`ADC1` peripheral. Before reading, the leads-off pins (`LO+` and `LO-`) are checked; if contact is lost, the read is skipped. This is done so we do not process floating inputs, which would trigger unwanted movement.
159
+
160
+
The signal is centered at around 8000 counts at rest, with 3.3V from Vcc, over the 14bit ADC.
161
+
```
162
+
midpoint = (1.65V / 3.3V) × 16383 ≈ 8000 counts
163
+
```
164
+
165
+
During contraction, the signal swings above and below that, with 10 samples taken, over 50ms. This deviation is checked against a threshold (trial and error), and the course of action is decided.
166
+
167
+
### Actuation
168
+
169
+
Detected contractions are mapped to three groups, reflecting forearm anatomy. The thumb and index have dedicated flexors, so each get their own sensor. The remaining three share the same muscle, as such, they share one sensor.
170
+
171
+
Servo commands are only sent on state changes (edge activated), one for flexing, one for closing. A 500ms debounce was also implemented, as to prevent servo chatter from the noisy EMG signal (please avoid these sensors).
172
+
173
+
The **PCA9685**, the **I2C** peripheral, drives all five servos, generating 50Hz PWM , independently from the MCU. The PWM frequency is set via a prescaler.
Each servo channel has four registers, `ON_LOW`, `ON_HIGH`, `OFF_LOW`, `OFF_HIGH`, encoding when the pulse starts and ends within the set cycle. The pulse always starts at 0, we only adjust the OFF value. By trial and error, we got to the following:
179
+
180
+
| Position | Counts |
181
+
|----------|--------|
182
+
| Finger open | 100 |
183
+
| Finger closed | 550 |
184
+
185
+
### Sensor Placement
186
+
187
+
The three sensors cover three anatomical regions of the forearm. The initial plan was to assign them based on known physiology, but as the sensors were too noisy, fine tuning was necessary. Anyhow, the muscles we sampled were the **Flexor Pollicis Longus** (for the thumb) and **Flexor Digitorum Profundus** (for the fingers).
188
+
189
+
I would insert a really nice photo here, but copyright laws forbid me from, and I do not intend on performing anatomy on myself. These can be easily seen via an internet search.
190
+
191
+
### Known Limitations
192
+
193
+
The **AD8232** is an ECG sensor, with a bandpass filter optimised for **0.5-40Hz**. Surface EMG, ideally, goes across **20-500Hz**. That means that we only capture the low-frequency, leading to forceful contractions. A dedicated module would improve sensitivity significantly, but the cost difference is substantial.
194
+
195
+
The threshold, of 3500 counts, was, as previously stated, determined via trial and error. Ideally, we would calibrate it at startup, per individual user.
196
+
197
+
The **PCA9685** was damaged during the final assembly due to a short on the power servo rail. Servo control was working fine prior to this, so a replacement module should restore full operation.
The Pass Vault is a hardware security device that stores encrypted passwords and types them automatically when a recognized fingerprint is scanned. By acting as a USB keyboard (HID), it requires no special software on the host PC. It features a Multi-Profile interface where each finger triggers a different credential, all managed via a secure USB-C smartphone or pc companion app.
16
+
17
+
## Motivation
18
+
19
+
I chose this project to create a physical-first security solution that eliminates the vulnerability of software-based password managers. It combines biometric authentication with Rust’s memory safety to ensure that credentials remain unhackable and accessible only by the physical owner.
20
+
21
+
## Architecture
22
+
23
+
24
+
## Log
25
+
26
+
<!-- write your progress here every week -->
27
+
28
+
### Week 5 - 11 May
29
+
* Bought and collected all the hardware components (Olimex RP2350 board, SFM-V1.7 sensor, and XIAO RP2040 board).
30
+
* Checked the pinouts and connection diagrams for the parts.
31
+
### Week 12 - 18 May
32
+
* Wrote most of the microcontroller firmware code in Rust.
33
+
* Studied and tested the `embassy` and `usb` libraries to understand how asynchronous hardware communication works.
34
+
* Made the fingerprint sequential enrollment logic and the USB keyboard emulation.
35
+
### Week 19 - 25 May
36
+
* Created the desktop companion application for the PC using egui.
37
+
* Designed the 3D case for the device and successfully 3D-printed the final physical enclosure.
38
+
## Hardware
39
+
40
+
The project uses the high-pin-count RP2350B, featuring 16MB FLASH and a 8MB PSRAM for advanced data handling and an XIAO RP2040 as the debugger.
41
+
42
+
### Schematics
43
+
44
+
45
+
46
+
47
+
48
+
49
+
50
+
### Bill of Materials
51
+
52
+
-->
53
+
54
+
| Device | Usage | Price |
55
+
|--------|--------|-------|
56
+
|[RP2350-PICO2-XXL OLIMEX](https://www.tme.eu/ro/details/rp2350-pico2-xxl/kituri-de-dezvoltare-altele/olimex/)| Main MCU |[55 RON](https://www.tme.eu/ro/)|
|[Seeed Studio XIAO RP2040](https://www.tme.eu/ro/details/seeed-102010428/kituri-de-dezvoltare-altele/seeed-studio/xiao-rp2040/)| Hardware SWD Debug Probe |[30 RON](https://www.tme.eu/ro/)|
59
+
60
+
61
+
## Software
62
+
63
+
| Library | Description | Usage |
64
+
|---------|-------------|-------|
65
+
|[embassy-rp](https://github.com/embassy-rs/embassy)| Async runtime and HAL for RP2040/RP2350 | Provides async drivers for hardware timers, UART communication with the sensor, and internal flash controllers. |
66
+
|[embassy-usb](https://github.com/embassy-rs/embassy)| Asynchronous USB device stack | Handles the concurrent USB stack, configuring the device as a composite USB CDC Serial class and USB HID Keyboard. |
67
+
|[usbd-hid](https://github.com/lyonel/usbd-hid)| USB HID class and report descriptor generator | Generates the official boot keyboard reports used to type decrypted characters into any OS host text-field. |
68
+
|[embedded-storage](https://github.com/rust-embedded/embedded-storage)| Storage traits for embedded systems | Provides standardized interfaces to safely interact with non-volatile memory sectors. |
69
+
|[eframe / egui](https://github.com/emilk/egui)| Immediate mode GUI framework for Desktop | Drives the minimalist PC companion application interface, managing background thread message passing channels. |
70
+
|[serialport-rs](https://github.com/serialport/serialport-rs)| Cross-platform serial port library for Rust | Enables the desktop client app to perform background plug-and-play USB hardware polling based on unique VID/PID signatures. |
71
+
72
+
## Links
73
+
74
+
<!-- Add a few links that inspired you and that you think you will use for your project -->
0 commit comments