Neurogress: a platform of neural control systems from Blue Brain project participants
The past 2017 year was a landmark for various startups who raised funds for their development through ICO. We witnessed the creation of a virtually new industry from the ideas of a handful of enthusiasts. Of course, there were both white and black sides in raising funds.
Today I want to tell Habr's readers about one project - Neurogress , which was recently announced by the team that worked in the Blue Brain Project . In short, the essence of this project is to create an ecosystem in which software could be developed for the neurocontrol of various devices (robotic prostheses, drones, objects from the Internet of Things, IoT, and so on) create the devices themselves and train the algorithms for the development of the Artificial Intelligence (AI) system of neural control (and even get money for it, tokens).
')
First of all, it will be of interest to those who are engaged in the field of artificial intelligence, pattern recognition and machine learning. On the other hand, everyone can support a project that will open a pre-TGE (Token Generation Event) the other day.
Intro
As White Paper (WP), published here , tells us, each of us is one big neural interface between our brain, the one who knows and exploits reality, and the often aggressive environment. Over the past thousand years, we have invented many pseudo-interfaces that help us adapt our mortal body, but for which we need hands, feet, and fingers.
The evolution of interfaces: from ancient times to the present day.Source: Neurogress Presentation
Now imagine that one fine moment of time such a “conductor” of the will of the brain will not (illness, accident, accident, and so on), a person will be limited in his actions and abilities. In this case, a replacement is necessary and it is desirable that it be controlled directly from the brain.
The Neurogress startup is designed to help develop such neuro-control systems that could directly transmit signals from the brain to robotic devices (for example, the same prostheses). The basis of the team are members of another high-profile project to conquer the brain - the Blue Brain Project (I wrote about it here and answered questions here ). Therefore, the project is based in Geneva, Switzerland, where the headquarters of the Blue Brain Project is located. The team decided to apply their accumulated knowledge and experience in the development of a neural control system based on artificial intelligence and machine learning.
The main goal of the project is the development of software for the neurocontrol of various electronics, devices and machines. In this project, neuromanagement implies patterns of brain activity (brain waves), which are collected using a neural interface and analyzed by machine learning methods and AI, and then transmitted as commands to a specific device. It is important to note that the project is aimed specifically at working with non-invasive signal reading devices, which should greatly simplify the introduction of technology, but more on that below.
An inquisitive reader will ask:so what about the mechanical / bionic limb and its direct interaction with the brain?
At the initial stage of the project, artificial limbs and mini-robots will be a “ model system ” for running solutions. However, the team plans to adapt its software for other areas: control of drones, machines and robots in production, new smart home systems and many other devices that could and, most importantly, should be controlled directly from the brain.
But let's deal with everything in order!
Briefly about neural interfaces and capabilities of neural control systems
The idea of ​​controlling technology using the power of thought has repeatedly appeared on the pages of science fiction and in games. Probably, many people remember how the cunning Yuri tried to seize power in the Red Alert 2 universe with the help of a peculiar neurointerface and psi waves.
From the vastness of the Internet
In reality, of course, everything is not so simple as in books and games. The best minds of mankind are struggling to enable us to manage reality with the power of thought. The brain is a very large computer in which electrical discharges run all the time (as long as the brain is alive) and, therefore, they can be counted and analyzed, revealing a “gentle” signal of brain activity among extraneous noise. Actually, this simple, at first glance, idea forms the foundation of all neural interfaces and control systems based on them.
Neuro management technologies can be divided into several groups:
Invasive. Electrodes should be implanted in the human body, most often in the brain. The main drawback of this type of technology, despite significant results achieved, is intervention in the internal environment of the body. As a result, the high cost of the error and the decrease in sensitivity with time (the body is rebuilt and "protected" from the intervention).
Miosensors. The electrodes are located on the human skin and read the impulses passing through the muscle fibers. Perhaps the most famous example, thanks to Stephen Hawking. However, this type has a significant delay: the signal from the brain through the central nervous system to reach the muscles and only then it will be read and interpreted, and, therefore, the delay is a few seconds.
Non-invasive. Based on various techniques for recording electrical activity of the brain using external devices. The main drawback is the non-specificity of the information received and its extremely complex interpretation, which requires a meticulous approach, the development of large data sets and the isolation from them of statistically confirmed information. In more detail about the disadvantages of non-invasive methods and the advantages of applying AI to this problem are described in WP .
The last type is the goal of the Neurogress team, which plans to create such a system for collecting data, processing and applying it. That non-invasive technology will create a product for the mass market, which entails a number of positive aspects. For example, the mass will allow (i) to attract more programmers to develop third-party applications, (ii) to identify problems in algorithms and learning models in the shortest possible time (below I will give an example of how it is planned to be implemented through NRG tools), (iii) reduce the cost of using software and , ultimately, devices with neural control systems.
An example of a device for non-invasive reading brain activity.Source: WP Neurogress
Neurogress Solution
The Neurogress team focuses on the development of algorithms and machine learning to recognize patterns of electrical brain activity and correlate them with limb motility. Since the AI ​​approaches will be applied, the final system will adapt to each specific customer with its features. It is planned that the learning process will take only 2-3 weeks to control the robotic arm instead of 4-18 months for existing solutions in the markets.
Here is a video demonstrating the capabilities of neuromanagement systems after some training:
The video shows that the system works with a delay. One of the goals of the project is to bring this delay to at least 0.5 seconds, and at best, to 0.05 c. In this case, the robotic limb itself can have up to 28 moving parts. And the cost of the product itself will be reduced from 2 to 5 times precisely due to a simpler and easier interface between the brain and the device.
Block diagram of the system.Source: Neurogress Presentation
How are you supposed to collect and process data?
Non-invasive brain activity sensor (neural interface) reads the electrical activity of neurons
Signals are recognized by frequency. In total there are 8 types of basic brain activity signals, combinations between which can be implemented for neurocontrol devices
Approximately 10% of the data remains unchanged, while 90% will be used to train the system with different patterns.
The results of multiple workouts are compared and the best solution is remembered by the system and then played back.
As soon as the signal is recognized and classified, the command is sent to the device to perform a specific action.
Thus, the software from the Neurogress team is a kind of intermediary connecting the neural interface itself and the robotic systems that the brain can now directly control. At the same time, the constantly accumulated data array allows to increase the accuracy and regularly update and modify the algorithms, drivers for devices that are under neuromanagement. As presented in the two diagrams below.
Integration of AI into neural control systems.Source: Neurogress presentation, also available in WP Neurogress
Inside the AI ​​learning ecosystem from Neurogress.Big brother is no longer following, but helps to adapt to new challenges.Source: Neurogress presentation, also available in WP Neurogress
Blockchain and NRG coins as the basis of the project
One of the features of the created platform and ecosystem is the presence of its own cryptocurrency - NRG (as long as these are tokens, but by the middle of 2019, that is, a year after TGE, the tokens will be turned into coins, a full-fledged cryptocurrency). It will not only be used to make purchases within the ecosystem and pay developers (as in the figure below), but it can be minted. Under the mining means the physical use of devices with solutions from Neurogress and AI training, as soon as successful recognition of brain signals is achieved with sufficient certainty, a coin is emitted.
Schematic movement of NRG coins in the Neurogress ecosystem.Source: Neurogress presentation, also available in WP Neurogress
Why blockchain? Modern giants such as Microsoft, Google, Facebook and many others “ monitor ” users, collect statistics and use it to improve and promote their products. When it comes to how often you use e-mail (gmail or outlook) or what you eat for breakfast (Instagram aka Facebook, Pinterest, Google Photos), then there is no problem that you are “ peeping ”. This is the inevitable consequence of the development of technology. However, when it comes to typical patterns of your brain activity, it is unlikely that users voluntarily agree to transfer these data to a particular company.
The Neurogress approach is that the development of AI is decentralized, each has access only to its own small piece of data, which it processes. At the same time, protection against monopolism over statistical data is implemented.
The savvy user will ask:there are a lot of popular crypto-currencies, such as ether (ETH).Why not base them?
Indeed, at the initial stage of NRG, tokens are tied to the airwaves, however, Neurogress plans to acquire its own currency in order to increase the speed of transactions and lay down its own mining principles for NRG coins. The latter will depend solely on how effectively Neurogress products are used.
Coins NRG and their use.Source: Neurogress Presentation
Neurogress Ecosystem
As already mentioned, the central part of the ecosystem will be the AI ​​and machine learning platform (AI Software in the diagram below), which will collect data, process it and turn it into signals for specific devices.
The main elements of the ecosystem Neurogress.Source: Neurogress presentation, also available in WP Neurogress
Also, it will include the following modules:
Neurogress Training Center. It is a virtual training camp, which is designed to help future customers hone their object management skills through neural interfaces. The main role of such a center is to ensure safe, stable and confident use of Neurogress software, be it a drone, a mechanical arm or a robot. Training will be carried out remotely using virtual reality technology.
Neurogress Hub. This platform is necessary to ensure communication between individual developers and project participants. Here, IT specialists and technology enthusiasts will share experiences, communicate and develop joint projects. Plus, it is planned that the Hub will organize meetings offline (meet-up in various areas of the company). Thus, the Hub will allow to support the development for the maximum possible number of devices.
Market development. It is the development market that will become the place where it will be possible to implement the lion’s share of NRG coins for the purchase of ready-made software, compatible hardware, algorithms and various services. A manufacturing company, a community of developers and partners will be grouped around each product. Equipment manufacturers, even if they use only a part of Neurogress solutions, will have full support from Neurogress (algorithms, user interaction, aka UX research, research and development, and so on). In addition, all proposed and released products will be fully categorized and available for easy search.
Stages of the project
At the initial, preparatory stage, Neurogress developed a beta version of the software for neuromanagement and tested it on a simple prosthetic arm and mini-robots, while collecting important statistical information.
At the next stage, which will last until the summer of 2018, the team plans to complete software testing cycles and submit a beta version of the prosthesis control for 8 movements. Already in May, after the completion of the initial round of TGE, a test version of the robotic arm and the software for managing drones will be released. Upon completion, TGE Neurogress plans to submit its own API and lay out the source code for open access.
The main stages of financing the project and the amount of work done with the funds raised.Source: Neurogress presentation, also available in WP Neurogress
Further - more, in 2019, an online platform will be launched, which will include several elements for interaction between customers, third-party programmers and software developers (more on this below), integration with IoT infrastructure, and the beginning of the full development of mechatronic prosthetic hands with an eye to collected statistics.
Business model
Neurogress's potential competitors in the emerging market for neuro-management devices are: the American company Cyber ​​Kinetic, the Australian Immotio, and the Russian startup from Skolkovo. At the moment, there is no accurate forecast and dry numbers for this market, but the authors of NeuroNet assume that various neurotechnologies will total about $ 1 trillion by 2035. At the same time, growth will be mainly provided by mass demand for augmentation (artificial ears, eyes, limbs), as well as neuromorphic computers and neural interfaces to control home devices.
In the short term, the Neurogress team plans monetization paths related to development in the following areas:
Software for neurocontrol systems
Joint projects on integration of neural interfaces for the control of mechanized limbs with industry leaders
Creation of neuro-editable humanoid robots and industrial robots together with leading manufacturers
Creation and development of the market of software developers and engineers
Creation and development of a market for training algorithms
Neurogress Ecosystem Development
Plus, the company plans to create a market where users can create and sell devices for people with disabilities, which will be based on software from Neurogress. Monetization will be carried out at the expense of deductions for the use of such software. At the same time, depending on the amount collected during the release of tokens, the company will put in the open access code that each developer can build for themselves.
To support user enthusiasm, the company will create a Neurogress ecosystem fund, which will invest 10% of the company's profit. Fund money will be used to support developers and for educational purposes. Monetization is achieved through the sale of training packages and tickets for seminars.
The main markets where Neurogress plans to enter (specific figures for each market are given in WP ):
Prosthetics
Robotic systems
Internet of things
Drones and unmanned aerial vehicles
Virtual Reality (AR / MR / VR)
Smart home systems
Instead of conclusion
Startup Neurogress sets itself extremely ambitious tasks that scientists have tried to solve for many decades, but only today with the development of AI systems and machine learning it becomes possible to translate into code and hardware. Plus, the development team will not have much time to "think." The development schedule is quite tight and will not tolerate delays.
Very soon, on February 10, 2018, a pre-TGE company will start, during which it is planned to collect about 15.7k ETH for the full funding of the project. It will last until March 25. The main round will begin on May 1, 2018 and last 2 months until the end of June. For this round it is planned to attract another 42.2k ETH. In total, the team intends to help out 58.5k ETH for the development of the project.
The project also has a Bounty Campaign promotion program with support from members via telegrams .
PS: as in the case of the Blue Brain Project, Habr's readers can ask all the questions of interest in the comments that will be given to the Neurogress team and to which they will try to give exhaustive answers.
PPS: Do not forget to subscribe to the blog and look at my Telegram channel : You are not difficult - I am pleased! And yes, about the defects noted in the text, please write in the LAN.