Mergen is a real-time artificial intelligence, which shall:
- serve as an operating system for any kind of robot.
- process and store vision, audio and touch as its input data.
- simulate the Reward System of the human brain.
- simulate both the Conscious and the Unconscious parts of the human brain.
- produce speech, movement commands and probably imagination as its output data.
Mergen is intended to become a robot which can think more logically and wisely than human beings, and potentially help them have better lives and consequently a better world(s)!
Emotions cloud human beings' judgements! Mergen won't be making mistakes the way human beings do out of their wishful thinking neither in its judgements nor its decision-making.
Mergen was named after a Turkic god of abundance and wisdom!
Mergen is being written in C++ 20 and is temporarily mounted on Android. The library located in the /cpp/ is intended to be mounted on real robots with a Unix-like operating system.
| Codename | Description |
|---|---|
| AUD | controlling Auditory inputs and outputs |
| HPT | controlling Touch (haptic) inputs |
| MOV | controlling Movements |
| REW | simulation of the Reward System |
| SCM | simulation of the Subconscious Mind |
| VIS | controlling Visual inputs (and outputs) |
At 21 August 2020, I wanted to programme a simple Android app based on logical theorems using which one can enter a sentence as a theorem and numerous chance factors below it which can make it True or False. The app had to calculate their mathematical mean and say, for example:
There is 20.4657% chance that a god exists!
This idea gradually turned into a robot who could do the exact same thing!
As I was getting more and more familiar with the machine learning concepts, I wanted to programme an Android app whose programming language was merely Python using Chaquopy, but since it had a weird bothersome license, I decided to forget about using Python in Android. Then I decided to try ML Kit, but it had no use to this project either! I wanted to name the robot I had planned to create "Horus", but then I found a better name for it!
The first version of Mergen, started at 13 December 2020, was intended to be a natural language processing (NLP) bot, something like ChatGPT today. In other words, it was going to simulate brain of a mature human being.
It was intended to process only digital texts in a server using pure Python, and the server had to communicate with an Android client app. But I concluded that such kind of robot has little "actual" use to human beings; therefore I cancelled its development and pursued a better method in Mergen II, see below.
The second version of Mergen, started at 10-12 January 2021, was intended to primarily process audio, I also called it the Pronouncer. It was a mixture of a Speech Recognition program and a Text-To-Speech (TTS) engine, which used International Phonetic Alphabet (IPA) to pronounce vowels and consonants, instead of normal alphabets. It synthesised human voices and was intended to be programmed to learn from voices and speak any kind of them. In other words, it tried to imitate a human child!
As I learnt more and more about Telecommunications, I preferred a more advanced method in Mergen III, see below.
The third version of Mergen, started at 29 March 2021 after many researches, had the same server-client model, this time "temporarily" on a Local Area Network (LAN), instead of a server thousands of kilometres away!
Since my smartphone couldn't handle such heavy jobs and my laptop couldn't move, I decided to do the heavy job of training/learning in a local PC server in Python and alongside an Android client in Kotlin whose...
- Camera served as an eye.
- Microphone served as ears.
- Speaker served as a mouth.
Mergen I was going to be like an adult human being and Mergen II was a human child, but I decided that I should start from something much, much earlier: an animal. I decided to programme an animal, then gradually raise it to be perhaps someone more logical than a human being!
Gradually I concluded that this server-client model and these radio waves take such a huge amount of time to be transferred and processed, and since I couldn't afford robotic resources, I was disappointed, so I cancelled this one too after a year at 10 March 2022 :(
I decided to use C++ instead of Python and do all the job temporarily inside an Android smartphone, so I began developing Mergen IV at 2 August 2022!
At 10 July 2023, I started a subproject for the computer vision (VIS) component and named it MyCV. It uses Python for faster debugging of the vision-related methods. After its successful image processing, I started translating it to C++ since 24 September 2023. Fortunately the C++ implementations worked in my phone, Samsung Galaxy A50, 10 to 50 times faster than Python(3.11) codes in my laptop with a CPU of Intel Core i7!!
At 16 October 2023, I started another subproject, called it MergenLinux, which runs Mergen on the Linux OS, so that I could learn how to implement them in a more difficult and lesser known OS, such as FreeBSD. But surprisingly, with few modifications it could run it on FreeBSD at 1 November!
At 3 November 2023, I started another subproject, this time for the audio processing/sound recognition (AUD) component (in Python again like MyCV) and named it Echo. Since Linux (Ubuntu) couldn't do the distance on my new decision for AUD to implement Open Sound System, I archived MergenLinux, and then forked it as Mergen4BSD at 12 November 2023. At 28 December 2023, concluded that creating subprojects for AUD and HPT is useless; because they don't need such pre-processing of their signals as VIS requires for its image frames!
After I realised that FreeBSD has so limited support for Vulkan, OpenCL and CUDA, I was disappointed with FreeBSD and archived Mergen4BSD at 21 April 2024! And decided to continue MergenLinux, but not so simultaneously.
At 30 April 2024, started another subproject called MyPR for Pattern Recognition.
At 27 May 2024, decided to change Mergen's way from being used on a phone to working in a 3D simulated environment, and decided to stop Mergen v0.4!
This table shows how much I've worked on Mergen and when. The timeline is divided into multiple consecutive periods. (updated as of 6403.03.07+)
| # | Starting date | Duration | Active days | % | Version |
|---|---|---|---|---|---|
| 1 | 2020/12/13, 6399/09/23 | 43 days | 20 | 47 | 0.1..0.2, pronouncer |
| 2 | 2021/01/25, 6399/11/06 | 264 days | ~33 | 13 | 0.3 |
| 3 | 2021/10/16, 6400/07/24 | 145 days | 18 | 12 | 0.3 |
| 4 | 2022/03/10, 6400/12/19 | 137 days | 0 | 0 | - |
| 5 | 2022/07/26, 6401/05/04 | 109 days | 7+8 | 14 | 0.4 |
| 6 | 2022/11/12, 6401/08/21 | 126 days | 13+27 | 32 | 0.4 |
| 7 | 2023/03/18, 6401/12/27 | 42 days | 0+1 | 2 | - |
| 8 | 2023/04/29, 6402/02/09 | 217 days | 88+13 | 47 | 0.4, mycv, linux+bsd |
| 9 | 2023/12/02, 6402/09/11 | 177 days | 64+1 | 37 | 0.4, bsd+linux, mypr |
- Periods 1, 2 and 3 include days of researches which are not recorded.
- Period 5 includes a few days of researching/practicing about/on a new programming language for Mergen (4 days), JNI+NDK (3 days) and OCR (1 day).
- Periods 6, 7 and 8 include a few days when I was learning/evaluating Vulkan (18-0-0 days), OpenBSD (2-0-0 days), FreeBSD (1-0-4 days), TensorFlow (3-0-0 days), PyTorch (3-1-0/5 days), Statistics (Data Science, 0-0-3/5 days) and Machine Learning (0-0-5 days).
- Period 9 includes 1 day for configuring clspv.
Main.java provides a basic control panel for this robot.
- Double-tap anywhere to start Mergen.
- Press the Back button of the phone to stop Mergen.
Once Mergen is started, it'll receive all kinds of visual, auditory and haptic inputs. Ergo, touching the screen while Mergen is online is handled by HPT as a haptic input; as if it senses your finger and processes your pattern of touching! That's why I didn't use double-click to stop Mergen.
There is also a remote control / debugging tool, bridge.py, which communicates over a network (like Wi-Fi) as a client with Bridge.java which is the server.
The app requires Android 10+. It uses Java 21 and the latest stable versions of Gradle, Android SDK, NDK and CMake. No third-party dependencies needed at all as of now.
Copyright © Mahdi Parastesh - USE IT WELL!!