This site lists some of the achievements of the past 25 years. Achievements are classified either by order of publication on this website, either chronologically or by topic. The transition from one system to another is via the tabs above.

Ambermind (consulting and technology creation) and other recreations

Cryo: the Second World and Scol

The student: Polytechnique, unix, MIT, and various hacks

Genesis: To7, Atari St

The first confinement was for me and for many others the occasion to take a step back and to see under a new light my own computer science career, started as a teenager by the discovery of basic and assembly in the 80's, then of C and Caml at École polytechnique.

I then realized how much, in 40 years, systems had gained in complexity and opacity while the gain for the user is not always obvious: boot time of several seconds, endless updates, fast obsolescence of machines, not particularly reactive user interface.

Even opensource does not escape this: a "new" linux server contains from the start about 200 processes, and thus as much attack surface, while nothing has been done with it yet.

For the developer, it is also a regression: the hell of dependencies, the stacking of black boxes, the instability of frameworks over time, the labyrinth of "support" forums, often outdated and contradictory.

This complexity and opacity are major obstacles when it comes to cybersecurity, sovereignty, and energy or material sobriety. The truth is that systems are out of control.

If we want to address these three subjects correctly, we must regain an understanding of the calculations actually performed by the machine. The "Less is more" principle was successfully applied 35 years ago in hardware to go from Cisc to Risc. It is time to apply it to software.

The solution I propose already has a history. In 1995, at the end of the X program, I had the opportunity to design a virtual machine and an associated programming language (functional language type) to create virtual worlds (Scol). The purpose of this virtual machine was to abstract and secure the execution of the code on the client.

In 2006, a much improved Metal version was integrated into the Nabaztag: the virtual machine also serves as the OS and Bios there.

In 2014, an even better Metal V2 version is used as the core of the downloadable version of 3D Slash.

In 2020, I started again from scratch with this time the objective to spread this technology, under the name "Minimacy" :

• opensource since the promise is to control the whole operation of the machine
• a unique combination of Virtual Machine + high-level language
• a Bios and a reduced set of essential libraries, written in this language
• a virtual machine that runs on Windows, Linux, RaspberryPi, MacOs, iOS, Android and soon directly on hardware.

By applying the "Less is more" principle, Minimacy reduces the complexity of computer systems to the size of a human brain, unleashes creativity, and opens the door to new solutions to problems of security, sobriety, and sovereignty. In doing so, Minimacy makes of computing a human thing.

To be part of this adventure, start here: https://minimacy.net/

Libsic Studio is a very personal work that brings together several topics that are important to me.

It is a work on the creative energy, the vital force that is the essence of Mankind. It reveals us to others and lets us accomplish ourselves.

Together we can lay the foundations of a more humane digital world, based on creativity, curiosity and goodwill.

Libsic Studio is a digital twin of the Earth, social and programmable. It is based on a large satellite image of our planet, broken down into small square Cells about 1 meter wide.

You can explore or redefine the appearance and behavior of each of these cells, either with the elementary bricks available or using Libsic, a programming language derived from Basic.

LIST, RUN, EDIT, PRINT, FOR ... NEXT, LOAD, POKE, ...!

Each cell can be seen as the 64x64 pixel screen of a small computer. This computer runs a program that you can modify as you wish. This program can react to different events, for example when you click on it or when your avatar passes through it.

For example, the following LIBSIC program displays a grid of red lines

In this world you are not alone, you share your creative energy with other users. Inspired by the slow web, Libsic Studio' social mechanisms do not encourage confrontation or provocation, but restore value to the contributions of individuals.

It is therefore also a work on Freedom: Freedom of creation, Freedom of expression.

This project combines several technical complications:

I hope you will enjoy discovering Libsic Studio and making it your own!

We have all already noticed the very particular aspect of the light point formed by a laser beam projected on a wall. There is a kind of grain that can't be found anywhere else. And yet this technology is little exploited: it is limited to pointers for lecturers, and to the visual effects of nightclubs.

There are obviously reasons for this, the main one being that we do not know how to deflect a laser beam other than by a mirror. In a vector cathode-ray display (oscilloscopes, Vectrex console) an electric field is used to deflect an electron beam, and this field can be modified instantly. With a laser beam, two rotating mirrors placed in galvanometers are used. The movement of the mirrors is therefore relatively slow and hampered by issues of inertia. In fact, only a few thousand positions can be transmitted per second, and only if the successive positions remain close.

Despite these constraints, and using a Lumax card, I was able to develop an Asteroid laser for the Cité des Sciences. The quality of the video below is quite poor, because the combined strobe effect of the camera and laser projector makes it difficult to have a stable image.

This summer one of my nephews came with a book full of Takuzu grids. Takuzu is a kind of Sudoku with only 1 or 0. There must be exactly five 0 and five 1 on each column and row, and there cannot be three identical neighbors.

As a summer recreation, I found it fun to write a program that automaticaly solves the grids from the book. The program uses the webcam to analyze the grid, then computes the solution and fill the grid directly on the webcam view, in real-time!

This may give you and idea of what you can get with an augmented reality device such as the coming Microsoft Hololens.

3D printing radically changes our relationship to the material. More than just a revolution of some industrial processes, it finally gives each individual a power over material: think about an object and get it for real. However for this revolution to be possible, an easy-to-use 3D design software is required, and there was none for real consumers. Try to open Solidworks or even Blender and you will understand the problem.

In addition, each of us is not so creative. The usual formulas like "you're only limited by your imagination" remind us that our imagination is very limited. While most 3D models available on the internet are trinkets designed by others and without much interest, regular people have simpler needs, especially in the field of repair or replacement. A useful tool must be able to create shapes with very precise dimensions because these shapes must fit with existing objects.

Watching children play Minecraft, I thought that there was something to explore: a very primitive action such as adding or removing a cube may be very powerful: look at what children realize in less than twenty minutes. However Minecraft model is too simple, it does not make sufficiently precise parts. A consumer 3D printer can typically print objects of size 10x10x10 cm, with a resolution of 0.1 mm, which makes 1 billion cubes!

Introducing several sizes of cube with a particular arrangement between cubes of different sizes, we obtain an intuitive and general purpose 3D design tool.

The result of this work is "3D Slash"

What it is:

• a web app with backup and sharing: www.3dslash.net
• a downloadable application made with metal technology, so that is will run on Windows, MacOs, Linux, iOS and Android. Also available on the website www.3dslash.net
• a file format "*.3dslash" which encodes your objects in an optimized way

This invention won the gold medal at "concours Lépine", the legendary French contest for inventors since 1901.

Go to www.3dslash.net!

Metal language developed in 2003 proved to be very useful in my work, either indirectly as a tool in prototyping or developing technologies, or as operating technology (see Nabaztag-tag ). However, some limitations have emerged, including the choice of C++, which significantly constrained portability. In 2003, the Windows PC was hegemonic except Linux servers. In 2012, it is necessary to also support Macs, iPhones, iPads and Android.

So I started to update the technology, while maintaining backward compatibility, and completely redeveloping with the C language, which is the only common point between these different platforms.

The syntax is slightly enriched: the definition of the constants is more flexible. Construction "for ... do" has grown. You can stop a function or a loop by a 'return' or 'break'.

Other improvements are :

• compatibility with Windows, Linux 32 or 64 bits, macOs, iOs, Android
• the compiled bytecode can be retrieved and used without the source code
• OpenGLES
• support of multi-touch and accelerometer
• support of utf8 and Json
• support of mysql client for desktop versions

The packaging is still work in progress, but you can try cross-platform on the example of 3D Slash:

I discovered the existence of 'Chocolate fix' looking for a gift for someone who loves chocolate. This is a puzzle published by Thinkfun. It involves placing 9 "chocolates" on a 3x3 grid in respect of a set of constraints, which make a "problem". Each constraint defines the relative position of chocolates, colors or shapes. In fact, the 9 chocolates are the products of three colors (milk chocolate, dark chocolate, strawberry) and three shapes (square, circle, triangle). A small book contains about 40 problems.

The game is very fun, but we quickly solve the proposed problems, and these are rather too simple.

The recreational idea is to automatically generate problems, including very hard problems. What algorithm? what complexity? It is required that the set of constraints is minimal, that is to say, it allows one and only one solution. I found a technique that generates three different levels of difficulty, depending on the type of constraints. On a processor Intel Core i7-2620M 2.7 GHz, used in single-threaded, generating problems takes between 1 and 6 seconds depending on difficulty.

The following interface allows you to look at these problems and their solution. Can you solve them? More importantly, how would you make the program that generates them?

In July 2011, the new owner of Nabaztag, Mindscape, filed for bankruptcy and announced the end of Nabaztag servers. There are still a few thousand rabbits alive, and there is something frustrating to see that the object has not been operated as it deserved. It presents an overall ergonomics of "answering machine": it receives messages, read them, and as it does not know if anyone was listening, it blinks. When you come home and the rabbit blinks, you know that a message has arrived and it can be read by pressing the button. The choice of receiving messages is done on a website, through complex settings.

Yet this rabbit was designed to have a real autonomy, and it has a fairly high power.

Having some free time at this point, I decided to set up an alternate server, which would explore a different way. This will be "Nabizdead", a neologism derived from a mixture of Nabaztag and Webisdead , referring to a party that I attended in early 2003.

This service offers a radically different approach: it is not necessary to go to the web site, everything can be done from the rabbit, using the button and ears.

The service works as a voice twitter: the rabbit picks in streams of short messages and reads them aloud. It invites to rate the stream (by raising or lowering the ears at the end of the message) which will influence the choice of the following messages. Similarly, it regularly offers to register to new streams. RFID tags can be associated with a stream.

The microphone is in fact always working, and it is the detection of a noise that triggers the reading of a new message by rabbits.

In addition, the rabbit gives the time, and retains its bedtime and wake-up time ; you put it to bed by lowering its ears.

By visiting the site you can create your own stream and post messages. It then displays the audience for these messages and the ratings from rabbits. An API can automatically post messages on these streams.

The service has been a nice success, considering the number of registered rabbits (over 3000 in early January 2012), and the number of streams created by users (more than 150, most of which automatically generate messages via the API to give the weather, news headlines, sport results)

The quick bankruptcy of Mindscape, followed by the buyout by Aldebaran, which reopened the official servers, cancelled the need for alternative servers, but it remains an interesting experiment.

Technically, the service opens new perspectives: hosting was considered too costly by Mindscape and is here replaced by a shared 60 euros per year server, which could easily handle the 10.000 rabbits that were still in use. The rabbit manages TTS by itself (text to speech), calling directly TTS services on the internet. To achieve this, it was only required to change the nominal bytecode.

Ergonomically, the service provides a community vision. According to the proven principle of Twitter, everyone easily generates content. The interaction is done directly with the rabbit taking advantage of its poor Input/Output in a very intuitive manner. Finally, the rabbit has a unique voice, there is no more "schizophonia".

Strolling randomly with Philippe Ulrich at CES 2007 in Las Vegas, we discover Koo Hyoung Lee on its booth (Neurosky) doing a demonstration of a helmet that can read your thoughts. We were working on interactive music , and it seemed obvious that the brain interface show great potential for this project because the music could automatically adapt to emotional state of the listener, or influence it.

I will run various experiments culminating in the completion of a working prototype (headphones + music software controlled by brainwaves) presented at Midem 2008. These helmets have the distinction of integrating both Brainwaves technology and audio headphones : as a happy coincidence, says Philippe, "ears are close to the brain, and it's probably not a coincidence." I asked Inprocess (who designed Nabaztag and Dal lamp ) to design the helmets. These helmets will win a prize of marketing innovation in 2008

Neurosky technology is integrated into a small electronic bluetooth module that uses three sensors: two are behind ears and get the ground reference voltage, the third sensor is located on the forehead and captures the brainwave.

In 2006, as part of my research on interactive music, I try to design a system that evolves randomly, but within predefined constraints and dynamic constraints, which are sometimes contradictory. This is actually a theoretical model for interactive works: the work evolves randomly, within the limits defined by the author and based on user actions.

I end up with a model that turns out to be a "Turing complete" calculus based on the notions of states and observables. The system computes probabilities for each state, and performs a random selection of the observed state (see quantum reduction). Probabilities are computed by predefined relationships in the form of probability contributions. Because it is about "mixing states," and it is a calculus, I will call it "mix-calculus."

After the success of the Nabaztag, the Violet company is developing a new and more powerful version with the same team: Philippe Debadier for electronics, and myself for the embedded software and the core of the server.

Little visible difference with its predecessor: a navel hides a microphone that allows the rabbit to hear and record sound. It also contains under its face a RFID sensor. However, the power improves with an ARM 6Mips, 128KB ROM for program and 1MB of RAM. To save in manufacturing costs, there is no flash memory and wireless card is SoftMAC: unlike the Nabaztag, Nabaztag-tag must implement in its firmware the entire wireless protocol (connection, encoding frames, ...). The SoftMac card processes mostly the radio part and encryption.

The embedded software is written in C. It manages various devices. It also manages the wireless card thanks to a driver written by a brilliant student, Sebastian Bourdeauducq I discovered by chance on the net in the credits of an opensource project for a Prism54 driver. Although he was still in selective studies, he realized the perfect driver.

Once again, it seemed natural to integrate a virtual machine to process the functional behavior of the object. First I attempted to integrate a Java VM, but I came up with a more controlled solution in terms of timeline, so I spent only a few weeks to integrate a simplified version of my VM Metal I had developed early 2003.

This virtual machine uses a "functional language", statically typed, providing great flexibility and with a garbage collector. Actually the network layer (arp, ip, udp, tcp, dns client, DHCP client and server, HTTP client and server) is implemented in this language.

At startup, the rabbit runs a first bytecode (called "boot" bytecode), which supports the connection to wifi. Once connected, the rabbit downloads the nominal bytecode ; this bytecode defines the behaviour of the rabbit. This operation is very open, even if the company Violet did not change the nominal bytecode much thereafter. The rabbit is particularly able to communicate directly with any server without going through the "main" server.

The bytecode is generated by a compiler, which is itself embedded in a simulator.

Late 2005 I met Fabien Vial and Pierre Varrod, who were starting a company about a tourism guide on your PDA / GPS that would be sold to famous Parisian hotels for their customers. This is an interactive map with touchscreen and lots of content (culture, shopping, restaurants, ...) in four languages ​​on a PDA with a GPS.

Not necessarily very original in 2012 at the time of the iPhone, but it is the end of 2005, and the product will be ready in June 2006.

I develop the app in C++ on Windows CE, and the target device is Fujitsu N520.

At the end of 2005 I am contacted by a small company that provides technical training material to schools. They are looking for a specialist of the motorola 6809 processor, and have heard about my To7 emulator which is also based on the 6809.

It is then decided to realize an electronic board based on 6809, with a Bios in rom, a serial link with a PC, and on this PC an application which serves as IDE.

The IDE allows to write the 6809 assembly language source code, the compiling to machine language, a debugger with breakpoints and step-by-step execution. The IDE also integrates an emulator of the board (based on the source code of my To7 emulator), and a serial link to the board to transfer the program in machine language and control its execution.

I believe that most computer enthusiasts have one day dreamed of building their own computer from scratch: this project gives me the opportunity to do it!!!

The most difficult part is to write the Rom of the computer. This one must be able to:

• initialize the system, the interrupts, the serial link with the PC
• provide basic functions to the user: hexadecimal display with 6 digits, keyboard reading
• communicate with the PC to receive programs and execute them
• provide debugging functionalities: pause/play, step-by-step execution, memory dump, display of registers values.

I start with the IDE and its integrated simulator, because I'll use it myself to write the ROM. In order to test it on the real board I use a ROM simulator, ELNEC SIMEPROM-01B, which looks like this.

In 2004, the Violet company stops commercializing the lamp DAL and wants to make a more general public connected object. After some hesitation between a flower and a teddy bear, the project ends up on a stylized white plastic rabbit. It will take a little more than one year before it is released in June 2005.

The electronic is designed ​​by Philippe Debadier, engineer at Sudria. It is based on an Microchip 8-bit microcontroller. I will take care of the embedded and server-side softwares. The system has 3.5 KB of RAM and 48KB of ROM to store the program, and 1MB of flash used as mass storage to store configuration and data received from the server. Its power is similar to the 8-bit computers in the early 80s, with a particularly small RAM.

The embedded software is written in C. It manages various devices and the wireless card in the PCMCIA format. The wireless card is of type 'HardMac', which means that most of the wireless protocol is handled in the card itself, including the connection protocol, encryption, frame encoding. RAM does not allow the complete storage of network frames, they are extracted (or transmitted) byte by byte from the PCMCIA card. Nevertheless, the firmware includes a relatively complete network layer: arp, ip, udp, tcp, dns client, DHCP client and server, HTTP client and server. As for the DAL lamp, the configuration is made in wifi adhoc mode through a small integrated http server, this mode is activated when you press the button on the rabbit startup.

The firmware also includes a rudimentary virtual machine, allowing it to receive bytecode from server, this allows to secure the proper functioning of the object, and also to change the internal behavior of the object after its commercial release. Actually, at the beginning of product development, we did not have a very clear idea about the services that would be offered.

This virtual machine uses a custom 8-bit assembly language which includes few comforts I met on the 16-bit Motorola 68000 processor (Macintosh, Atari ST, Amiga): there are 16 registers. Some opcodes are used to control the LEDs, motors, optical sensors and the audio system. The code name of this language is VASM.

Once connected to wifi, the rabbit downloads a first bytecode and regularly polls the server through http, and possibly receives a new bytecode, which will replace the previous one.

The source code for embedded software has not been released yet by the new owners, so we do not find it here.

The bytecode is generated by a java server that relies on templates of VASM sources, which are filled on the fly, and then assembled into the bytecode.

In early 2004, as the Nabaztag project begins , I try to upgrade myself in electronics by making a prototype of a dance mat for any dvd player, in the spirit of "Dance Dance revolution".

The idea is to connect the dance mat on any DVD player, and provide a DVD containing clips with dance steps.

Problem: there is no standard input-output on these DVD players. The solution is to use the only connector that is on all DVD players : the RCA video connector (yellow plug). This jack is only video output. To generate interactivity, we must find workarounds (like the old Teletex pages, which gave the impression of an interaction with a remote server).

The principle is as follows: DVD screens integrate drawings that look like black and white mosaics and contain a code. The electronic receives the video signal, recognizes these drawings, and extracts the code. For example this code contains a list of dance steps that the player must perform.

The electronic device then manages itself the game (by emitting sounds to indicate if the step is correct or not), and displays the score on a little digit.

I will complete a working prototype in wrapping, based on a Microchip 8-bit microcontroller (similar to the Nabaztag). Click on pictures to enlarge.

The project also includes a utility to define the steps and generate video with the scrolling arrows. This utility is developed with Metal

and a test DVD containing two clips, and a configuration management.

In 2003, I attended entirely by chance at a performance of Michel Risse, "Instrument | Monument" in the old station in Longueville (France) at Heritage Days :

This video shows the concepts, but unfortunately not at all makes the sound impression of the show, which was intense and gripping. In the game of musicians / dancers / acrobats, I see an interactive dimension (which is perhaps a wrong impression), where the music is simply the product of random interactions between these musicians. There seems to be too many uncertainties in the duration of acrobatics for the music to be expected, and however everything matches. The end comes as a fly in the ointment, frustrating: the music could continue indefinitely in a constant renewal.

These are the basics of what I mean by "interactive music": music that is generated in relation with external events, the duration of which is also decided interactively. This interaction does not prevent the artist from defining a framework of sounds, a melody, ... All this creates for the listener an engaging and entertaining experience.

You find below a very simple prototype of automatic and endless generation, performed by dynamic mixing of samples. The interaction is reduced to a minimum, the listener can just stop the music whenever he wants. The mixing is done on the client, the server merely provides samples as a simple http server.

After some experimentation in 2004-2005, the topic interests Philippe Ulrich, my longtime partner, and together we will create (with Gilles Babinet) Mxp4 (aka. Musinaut) that will explore this topic for two years before taking an unrelated direction. We will do various experiments based on popular songs interacting with the listener to choose the style and duration. We also test the control of music by motion sensors, accelerometers, and brainwaves.

Technically, this research will lead to the creation of the Mxp4 format, which allows you to define an interactive music based on a dynamic mix of samples. Choice of samples is the result of rules provided by the artist and of external interactions. The Mxp4 player exists in different forms: streaming with Flash player, offline on Windows, Mac, Linux, Windows CE, Symbian, iPhone.

The video above shows the iPhone version on a cover of Oxygene4 (Jean-Michel Jarre). The interaction is to choose styles of interpretation, each generating an infinite and different music every time you are listening. The "lock" button will lock the style, while in normal mode, the style will automatically change after a while. Although unpredictable, the music is accompanied by contextual comments on the instrumentation or the original song ...

Beyond the technical aspects, this research shows the difficulty of giving meaning to the musical interactivity ; if I move, the music changes, ok, but how to have it make sense? It is this ability to stage an interesting musical interactivity that opens up a new space for music creation. Who are these new artists? The project will demonstrate that, as a good photographer will not be a good filmmaker (though we are still talking about pictures), a traditional musician will not necessarily be an interactive musician ...

The first barrier is mental and is related to the completeness of listening. An author of video game easily claims paternity of his game, even if he cannot check all the possible games because they depend on the player behaviour. But a musician of the early 21st century wants to hear everything before putting his name on it.

The second barrier is precisely the ability to create an interactive musical space that makes sense, and this ability has no connection with the qualities of a traditional musician or composer. However a minimum of musical knowledge seems necessary.

In 2003 I was put in touch with Olivier Mevel, who is about to create the company Violet about pervasive objects and ambient intelligence. I will help for 3 years as co-founder and external CTO, taking care of all software issues. This is the fun of this kind of startup: for over one year, I will do all the technology. Violet then gradually internalizes web and servers aspects.

The first object is the lamp DAL. Why a lamp? because "light is life" as the designer told me. This lamp will be developed in less than 4 months. Its core electronic is based on Linux. It communicates with a Java server. To ensure the integrity of the operation, the onboard software is a virtual machine that downloads apps from the server and these apps run simultaneously in the lamp. The language for these apps is a specific XML language, which allows you to make simple computations and to access peripherals: 9 RGB LEDs and touch sensors.

The object is connected to data sources (weather, email, exchange, ...) ; the apps rely on these sources to generate light animations.

The object is connected through wifi. Configuration is made in adhoc mode through a small integrated http server (with a switch under the lamp to select the adhoc mode).

The server allows you to configure the services, but also to upload your own apps.

The lamp DAL will be presented in 2003 at the Pompidou, at Batimat, then at Apple Expo.

A video from Pierre Dandumont :

The Scol programming language I had developed in Cryo needed some improvements that it was difficult to integrate within a "backward compatibility". But after the end of Cryo I had free time and I could write a new and simpler version including incremental garbage collector, weak types, parameterized types, more efficient type unification, lighter syntax, ...

Bit by bit this project originally designed to "free my mind" has become useful when I started my consulting business. First considered as a prototyping tool, it has proved very useful, and a simplified version was even built in the Nabaztag-tag in 2006.

To concatenate two lists of any type, conc function will be defined like this :

The virtual machine contains graphics libraries, network, audio and openGL.

The virtual machine is available on Windows, Linux and MacOS.

In 2001, I have the opportunity to try the development of mobile games. It is the very beginning of this : the java standard for mobile games (j2me midp2.0) does not exist : the only available technology is called Exen (designed by the french company In-Fusio). It relies on Java and extends it with graphical library. The mobile screens are purely monochrome (without gray) and this will be painfull for the young graphist who didn't even know that this used to be like this on computers.

I decide to make a "tennis" game, because it looks appropriate to the graphics capabilities and the available cpu power. The position of the ball and its shadow are computed in 3d, and the AI of thie opponent requires some iterations...

The Exen system relies on Java, but does not benefit from its memory automatic management. You must configure the software by describing how the memory is organized (for example 16 blocks of 8 bytes, ...).

The game will be published by In-Fusio.

In 1996, as the Second World project is painfully almost released, we feel the need for a "middleware" dedicated to virtual worlds if we want to go further in this area. We start then a R&D project that will lead to the creation of the Cryonetworks subsidiary and the developement of the Scol technology.

The first presentation of the technology in early 1998 shows a world called Cryopolis and designed by Laurent Doubre.

There are two sides in this technology :

• a virtual machine dedicated to networked multimedia applications, based on the Scol programming language, which runs on both server and client sides. Influenced by my studies at MIT and Inria, the Scol programming language presents some similarities with functional languages such as Id or Caml, with a simplified syntax so that it is more familiar to traditional developpers.
• a research on distribution topics : each functionality is packaged into a module that contains the server code and the client code. In this "vertical" way, we can provide a graphical integration tool (the SCS) in which modules are linked together graphically by a user who knows nothing about hard distribution issues. This research leads to the "user-flow" concept, which will be published at the international conference GT-RV99 in Laval.

Even the Fnac (a famous french store) used it in 2000, five years before the "second life" buzz :

Cryopolis in 2000, a virtual community with 1.000 DAU, all of them connected with 56K modems :

Cryonics, a simplified tool to create virtual worlds. The world is hosted locally, on the user's PC (that you should not swith off) ; Scol, pioneer of P2P ?

A presentation of Cryonetworks in 2000, it is then a public company with about 180 employees ; the video presents online games (Venise and Mankind) and Cryopolis :

After the end of Cryo in 2002, I got back the rights on the Scol technology and I put it in opensource.

For the Imagina 97 exhibition, Philippe Ulrich and myself write a prospective paper on the mutations induced by the digital age.

It is all about information, copy, ubiquity ; and even lambda calculus in some way. At the end we annonce the Magma experiment which is the initial name of the Scol technology.

The Menta, fusion of information and computing, moving and ubiquitous, is still a subject of prospective.

In 1995 the new company of Philippe Ulrich, Cryo Interactive, asks me to help in a revolutionary project that connects two topics which were "hot" by that time : 3d and networks. It will be one of the first 3d virtual worlds. It is called Second World: a virtual Paris in which people interact through avatars.

I will be the CTO of the project and provide myself two components of this "game" : the 3d engine and the server.

The 3d engine is purely software (there was no 3d hardware in 1995) and must handle large outdoor areas with a high number of polygones; this was the opposite of the games at that time such as Descent or Quake : indoor and few polygones. Written in x86 assembly language, based on the "scanline" algorithm, it was using an eccentric technique to compute the texture mapping. Texture mapping requires theoriticaly two divisions per pixel, which is too slow in 1995 for real-time 3d. The trick is to compute on logarithms (through conversion tables) and then to replace these division by substractions.

A first version will be released in 1997, and there will be shortly a V2.

The server is running on Unix and can handle a few hundreds of simultaneous users. In 1995 internet isn't still very popular for the general public, and the game is initially designed for a private network of modems. But before the end of the production, the general public is rushing into IAPs and the game will eventually use internet.

We can navigate today into the 3D scenes of the Second World thanks to a prototype of virtual world based on Scol technology, and that should have been the core of a V3.

Early 2014 I restarted a server so that you can replay Second World on Windows, Mac or Linux computer, based on VirtualBox. The procedure is like this:

• download and install VirtualBox : https://www.virtualbox.org/wiki/Downloads
• download the Second world "virtual application" : http://dev.ambermind.com/2m/2M-virtualbox.ova
• launch VirtualBox
• in File menu, select "import a virtual application" and open the file 2M-virtualbox.ova you have just downloaded
• start the virtual machine "Windows 98 SE - 2eMonde" (windows 98 is starting)
• on the dialog "saisie du mot de passe réseau", leave empty and click OK or Cancel
• double-click on "Raccourci 2monde.exe"
• you need a user account to connect: send me a mail at lui-meme@sylvain-huet.com, telling me the login you would like
• I will give you a password ASAP

few hints:

• when using VirtualBox it is sometimes tricky to quit the virtual machine: on PC use right Ctrl to get the control. On Mac use left Cmd
• to quit the application it is preferable to quit properly windows 98. Menu Start, select Stop. The screen will then freeze on the picture "Fermeture de Windows". You may then close the "Virtual Box" window and select "shutdown".

In 1996 I have the opportunity to finish my studies at Inria (french famous computer sciences laboratory) in the PARA team lead by Jean-Jacques Lévy, on the topic "Security and mobile processus". This is the opportunity to discover the pi-calculus (a theoritical calculus such as the lambda-calculus, dedicated to distributed computing), and the join-calculus implemented by PARA and which allows the processus to move as they are running. The internship leads me to study cryptographic and right management in this very theoritical context.

In 1994, I have the opportunity to spend a few months at MIT, in Pr.Arvind's laboratory, Computation Structures Group. This team has invented "Monsoon", a Dataflow computer in association with Motorola. Dataflow is a very efficient parallel programming model : "Imagine a computer program that can run on thousands of processors with no modifications and even greater performance. Scalability and programmability would be virtually unlimited".

My work was to bring interactivity into this machine, which was missing it. I added X11 interface, and it was then possible to explore new applications such as the parallel computation of a spreadsheet. I also implemented what was probably the first parallel Breakout of the History.

The computer was using a functional language called "Id", very smart, very efficient, and very well-fit for parallelism.

Begining of 1994, I discover the "emulators", softwares that transform today computers into computers of the 80's. Apple ][, Commodore 64, Zx Spectrum, most of these old computers are already emulated. All ? no, because the essential french computer, the To7, struggles, probably because it is not known outside of France. I work on it and make my first computer revive. This is also my first "virtual machine" ; I do not know then that there will be a lot more later.

The first version is made for Unix/X11 (Linux, but not only). The ms-dos version will come in 1995, and the windows and pocket pc in 2000.

The essential tool for any emulator is the software recovery tool. For the To7 it is the tool that can retrieve the original softwares which were written on audio tapes. You first digitize these tapes (44.100Hz, 8 bits, mono), and then a software must analyze the sound and distinguish the sounds that encode a '0' and those that encode a '1'. Then if you know the encoding scheme of the To7 you can extract the complete information. In theory this looks simple but the tapes are not always well preserved, and errors will occur. How to find them ? The best method I experimented was to implement two separate tools based on two very different technics (for example one is computing on frequency space, one is computing on temporal space) : when an error occurs, both technics usually provide a different interpretation of the bad magnetic signal, which betraies its position.

A specific version of this emulator has been implemented for the French National Library (BNF), as this library conserves softwares, and especially a famous collection of educative softwares published in 1984-1986 when the french government bought Thomson computers in every schools.

In 1994, as 3d hardware does not exist, I am looking for algorithms to optimize the display of 3d objects. I imagine a method that pre-computes the drawing order of polygones, and the theoritical analysis looks interesting to me.

When an object is convex you can sort the polygones easily : first you draw the polygones with the inner side visible, then the polygones with the outer side visible.

When a plane divides an object in two sets of polygones, we display first the polygones which are not on the same half-space as the observer, and then the polygones which are in the same half-space.

With these two observations, we demonstrate that we can always decompose an object as a tree of these two cases. And unexpectedly some objects that look very complex can be decomposed with only the first case (such as a stairway).

As we arrived at Polytechnique in 1992, we were asked to buy an Apple laptop (powerbook 145), on which we would have to implement a 'computing project' on a subject we could choose by ourselves. As I was interested into operating systems, preemptive multitasking and graphical interfaces, I implemented such a system on Mac. A that time usual systems (Mac, Windows 3.1) were not preemptive : the window with the focus was the only one to do computation ; others were frozen. Applications had to give back the cpu to the system as they had finished their computation, else they would freeze the system. In a preemptive system the cpu is available successively for the applications and only for a short slice of time.

This project is built from scratch (it only gets the video memory address) : a monochrome graphical environment with windows and mouse and a task manager (with a schedulaer and multi-threading).

Several demos are integrated : the computation of Mandelbrot set (which demonstrates visually that many windows are working at the same time), a othello, a monitor of processus and memory, and some demos of concurrent programming (the classical "parking", with only one entrance, one exit and many cars), ... and a cheatmode, which is launched by 5-click.

The following screenshots have been made on Mess emulator for Mac SE/30 (because a 68030 is required as well as a monochrome display). The monitor of the Powerbook 145 was slightly bigger (640x400 instead of 512x384), therefore some pictures may look clipped.

Pacman is the essential reference, the archetypal video game, even if I never enjoyed the arcade version ; I prefer a more complex maze, with longer games such as on VCS2600.

In 1991 I implemented a personal version on Atari ST, for monochrome screens ; I implemented it later on various platforms, but it is always the same gameplay and the same mazes. This is my "hello world" favorite program. You can download all the versions.

The very first in 1991 on Atari ST monochrome, in 68000 assembly language:

Still on ST and 68000 assembly language in 1992, but with colours :

In 1992, Powerboook 145 version provided as a cheat mode in my Computing project, a mix of 68000 assembly language and C.

In 1993, Unix X11 in C language:

In 2003, implementation for my Metal language:

In 2012, Unix text version (ncurses library) in C language:

Last version, renamed as Dad-Man, a version for iPad, Android, windows, mac, linux written with Metal V2, based on openGL

In 1988, Ere Informatique asks me to adapt one of its successfull games, Macadam Bumper (by Rémi Herbulot) on Thomson computers. With very few documentation on the original software, this is a very free adaptation. It is a customizable flipper.

It is a mix of Basic and 6809 Assembly language, with very few Basic ; it is quite difficult to extract the Basic source code, except if you load it on an emulator. It was released as a 3'1/2 floppy

Just after the developement of Reverse, in 1987, I imagine a new solving algorithm, very simple, which will prove different and better than the well-known minimax. I will implement it first on Atari ST in 1990, but it is only in 1994 that I will have sufficient mathematical knowledge to analyse it.

In this kind of problem, we use an 'evaluation function' that associates a score to a game position. For example for a 5 half-moves search, we are looking for the move that secures the higher value of the evaluation function 5 half-moves later, whatever the opponent will play. The state of the art algorithm is called "Minimax", and has an optimized variant "Alpha-Beta".

In the "Penultimate" algorithm, we use dichotomy : we choose a value of the evaluation function, for example V, and we wonder if there is a move that secures this value. For n half-moves, we play randomly a fist half-move, then randomly the second, and so on until the n-th half-move. We consider then the obtained position. If its score is higher to V, we modify the last move of the opponent ; if no move is remaining, we go back by two half-moves so that we modify the penultimate move of the opponent, and so on. As we go back by two half-moves, we may go back before the current position : it means that V is secured, and that the first half-move of the player that is currently tested is the good one. If at some point we have tried all the first half-moves of the player, it means that V cannot be secured. By dichotomy on V, we determine precisely the best possible move and its associated V value.

All this is explained in detail in the following paper.

In 1986, the great french magazine "Jeux & Strategies" offers a puzzle based on Othello materials. It's basically to return all pieces from the board by a succession of moves (complete rule below). A free subscription is the winning prize. Before I had time to design the software on my TO7 computer, another player has found the solution without computing help. But the magazine repeats the challenge in three variants, and this time my software is ready.

On a TO7, it takes 20 minutes calculation to find the solution of the first problem.

With a little more experience (I was 14 in 1986), a new version written in 2012 solves the first problem in 15 seconds. To monitor the program without slowing down, it runs directly into the video memory, which unlike machines today is part of the main memory. This shows pixels "blinking" as long as the solution has not been found.

This first software is made ​​in 1986, at the age of 14, and spotted by Philippe Ulrich, who is then creative director of Ere Informatique, a French publisher that relies on a network of independent authors. It is published under the label 'Gasoline Software'.

This program is written for Thomson (To7, Mo5) computers, for which I later wrote emulators. It is a mixture of Basic and 6809 Assembly language. It was distributed in the form of audio tape.