Neural-Controlled-robotic-ARM-With-Kernal

About

• Low-level EMG robotic arm controller written in pure C/ASM signal capture, filtering, and actuation from memory addresses up
• Youtube Demo https://www.youtube.com/watch?v=1ESYieNTGVo
• Web Docs https://inborastudio.github.io/Neural-Controlled-robotic-ARM-With-Kernal

---

Hardware Summary

Component	Function	Key Parameters
EMG Electrodes	Signal Source	Surface electrodes detecting muscle potentials
Instrumentation Amplifier	Signal Conditioning	Gain 1000×, Band 20–450 Hz
ATmega328P	Core Processor	10-bit ADC, 16 MHz clock
Servo Motor	Output Actuator	PWM control at 50 Hz, 1–2 ms pulse width
Power Supply	Energy Source	5V DC regulated

---

Memory and Register Mapping

Microcontroller Memory Layout

Section	Address Range	Description
Flash (.text)	0x0000 – 0x7FFF	Program memory
SRAM (.data + .bss)	0x0100 – 0x08FF	Variables, buffers
EEPROM	0x0000 – 0x03FF	Calibration constants
Stack	0x08FF → 0x0100	Grows downward

Key Register Map (ATmega328P)

Register	Address	Description
ADMUX	0x7C	ADC Multiplexer Selection
ADCSRA	0x7A	ADC Control and Status Register
ADCH / ADCL	0x79 / 0x78	ADC Data High / Low
DDRB	0x24	Data Direction Register (PWM Output)
PORTB	0x25	Port B Output Register
TCCR1A	0x80	Timer/Counter1 Control A
TCCR1B	0x81	Timer/Counter1 Control B
ICR1	0x86–0x87	Input Capture Register (TOP for PWM)
OCR1A	0x88–0x89	Output Compare Register for Servo Pulse

---

Signal Processing Mathematics

1. EMG Signal Model

The raw EMG voltage signal can be approximated as a stochastic process:

[ x(t) = A(t) \cdot \cos(2\pi f_c t + \phi) + n(t) ]

where

• ( A(t) ) = envelope of muscle activity
• ( f_c ) = carrier frequency (50–150 Hz)
• ( n(t) ) = noise (thermal + motion artifacts)

---

2. ADC Quantization

For a 10-bit ADC at 5 V reference:

[ Q = \frac{V_{ref}}{2^{10}} = \frac{5}{1024} = 4.883 \text{ mV/step} ]

[ D_{out} = \frac{V_{in}}{V_{ref}} \times 1023 ]

Example:
Input 2.45 V → ( D_{out} = (2.45/5) \times 1023 ≈ 501 )

---

3. Full-Wave Rectification

Since EMG is bipolar, a full-wave rectifier converts all signals positive:

[ y(t) = |x(t) - V_{bias}| ]

Bias voltage: ( V_{bias} = \frac{V_{ref}}{2} = 2.5 \text{ V} )

---

4. Moving Average Envelope Detection

A discrete moving average filter of window length ( N ) smooths the rectified signal:

[ E[n] = \frac{1}{N} \sum_{k=0}^{N-1} |x[n-k] - 512| ]

Implementation (fixed-point arithmetic):

E = E + rect - buffer[idx];
buffer[idx] = rect;
idx = (idx + 1) & (N-1);
output = E >> 5; // Divide by 32

5. Mapping to Servo PWM

Servo pulse width varies between 1 ms (0°) and 2 ms (180°).

Timer1 configured at Fast PWM, prescaler 8:

This maps envelope strength linearly to servo angle.

Numerical Example

Input Voltage (V)	ADC Value	Filtered Envelope	Pulse Width (µs)	Servo Angle (°)
0.5	102	10	1000	0
1.2	245	80	1200	36
2.0	409	160	1400	72
3.2	655	250	1600	108
4.0	819	320	1800	144
5.0	1023	400	2000	180

ADC Timing Analysis

• For a prescaler of 128:

• Each conversion requires 13 cycles:

• Effective filtered sample rate (after averaging): ~1 kHz.

  | Stage | Approx. Delay | Notes | | ---------------------------- | ------------- | -------------------------------- | | ADC Conversion | 104 µs | Hardware | | Filter Window (N=32 @ 1 kHz) | 32 ms | Envelope delay | | PWM Output Update | < 1 ms | Timer immediate | | Total | ≈ 33 ms | End-to-end muscle → motion delay |

Variable	Size (bytes)	Address Range
ADC Buffer	64	0x0200 – 0x023F
Envelope Sum	2	0x0240 – 0x0241
Filter Index	1	0x0242
PWM Command	2	0x0243 – 0x0244
Stack	64	0x08BF – 0x08FF

Compilation & Flash

• Build with avr-gcc:

avr-gcc -mmcu=atmega328p -Os -DF_CPU=16000000UL main.c -o emg_arm.elf avr-objcopy -O ihex emg_arm.elf emg_arm.hex avrdude -c arduino -p m328p -P COM3 -b 115200 -U flash:w:emg_arm.hex

Safety and Isolation

• Never connect electrodes directly to the MCU input.
• Always isolate EMG amplifiers from the microcontroller power rail.
• Use differential amplifiers and a DC blocking capacitor on input stages.
• Recommended isolation IC: AD620 or INA128 with ±9V rails and 1 MΩ input impedance.

Credits

• Kernel Design by Seo Park Jun
• System Staging by Chethan Yadav
• Core System Designer Ofc me