<dt-article> <h1>Learning Sensorimotor Agency <br> in Cellular Automata</h1> <h2 id='finding-robust-selforganizing-agents-with-gradient-descent-and-curriculum-learning-individuality-selfmaintenance-and-sensorimotricity-within-a-cellular-automaton-environment'>Finding robust self-organizing "agents" with gradient descent and curriculum learning: individuality, self-maintenance and sensori-motricity within a cellular automaton environment</h2> <div class="l-page"> <video id="robust" width="95%" autoplay controls="" loop muted="" class="videoShow"> <source src="public/demo.mp4" type="video/mp4"> </video> </div> <h2 id='interactive-demo'>Interactive Demo</h2> <d-content style="padding-top:0"> <nav style="margin-left:2em;"> <p><a href="https://colab.research.google.com/drive/11mYwphZ8I4aur8KuHRR1HEg6ST5TI0RW#scrollTo=l6D-g1Q38yyC" class="colab-root">Reproduce in a <span class="colab-span">Notebook</span></a></p> </nav> <nav id="infoDemoAdvanced" style="margin-left:2em;display:none"> <li> <b>Press play to begin</b> </li> <li>Place creature by clicking on the dark screen with <b>Creature</b> mode selected)</li> <li>(Note that the multi creature setting was not seen during training)</li> <li>Now place obstacles by selecting <b>Wall dot</b> and then clicking or dragging on the screen </li> <li>You can erase by selecting the <b>Eraser</b> mode and then clicking or dragging on the screen </li> <li> You can choose the brush size to draw bigger obstacles/ clear bigger areas etc</li> <li> To zoom, select zoom mode, select zoom magnitude with the slider and then click where you want to zoom<br>Keep dragging to follow the creature </li> <li> For the attractor, click and drag for a short distance attractor (Still experimental, works best with creature 2) </li> <li> You can upload images through browse, We advise you to use small images (as big ones will make large grid that are computationally heavy)</li> </nav> <nav id="infoDemo" style="margin-left:2em;display:block"> <li> The demo displays a differentiable continuous cellular automaton (CA) system where an emerging "creature" (yellow) is the result of local rules applied on the whole grid. The blue elements are obstacles (and are also part of the CA system).The rules leading to these very robust creatures were obtained using gradient descent,diversity search and curriculum learning. </li> <li> <b>Press play to begin</b> (The demo won't work on phone) </li> <li>Choose the type of creature on the left (each creature morphology is the result of different rules in the CA)</li> <li>Choose pre defined environments at the bottom </li> <li>Draw obstacles by clicking on the canvas </li> <li>More options available at Advanced Options </li> </nav> </d-content> <div class="radio-toolbar" id="radioCrea" style="float:left;margin-left:calc(50% - 1284px/2);width:50px"> <input type="radio" id="demoButton1" name="optionCrea" value="1" checked> <label for="demoButton1" id="labelDemoButton1" onclick="setSpecies(0);document.getElementById('getStarted1').style.display='block';document.getElementById('getStarted2').style.display='none';document.getElementById('getStarted3').style.display='none'"><img style="width:80px;height:56px" src="public/crea1.png"></label><br> <input type="radio" id="demoButton2" name="optionCrea" value="2" > <label for="demoButton2" id="labelDemoButton2" onclick="setSpecies(1);document.getElementById('getStarted1').style.display='none';document.getElementById('getStarted2').style.display='block';document.getElementById('getStarted3').style.display='none'"><img style="width:80px;height:56px" src="public/crea2.png"></label><br> <input type="radio" id="demoButton4" name="optionCrea" value="4"> <label for="demoButton4" id="labelDemoButton4" onclick="setSpecies(3);document.getElementById('getStarted1').style.display='none';document.getElementById('getStarted2').style.display='none';document.getElementById('getStarted3').style.display='block'"><img style="width:80px;height:56px" src="public/crea4.png"></label> </div> <div class='l-body'> <canvas id="glCanvas" width="640" height="360" style="background-color:#000000;"></canvas> <!--shadertoy at mac 840x472, at win 640x360--> <div id="showText" onclick="this.style.display='none';">initializing...</div> </div> <div> <span id="play-pause"> <svg class="icon" id="play" style="display: inline;"><svg id="playIcon" viewBox="0 0 24 24"><path d="M8 5v14l11-7z"></path><path d="M0 0h24v24H0z" fill="none"></path></svg></svg> <svg class="icon" id="pause" style="display: none;"> <svg id="pauseIcon" viewBox="0 0 24 24"><path d="M6 19h4V5H6v14zm8-14v14h4V5h-4z"></path><path d="M0 0h24v24H0z" fill="none"></path></svg> </svg> </span> <svg class="icon" id="reset" style="display:none"><svg id="resetIcon" viewBox="0 0 24 24"><path d="M0 0h24v24H0z" fill="none"></path><path d="M12 5V1L7 6l5 5V7c3.31 0 6 2.69 6 6s-2.69 6-6 6-6-2.69-6-6H4c0 4.42 3.58 8 8 8s8-3.58 8-8-3.58-8-8-8z"></path></svg></svg> <div class="radio-toolbar" id="radioStart" > <input type="radio" id="startButton1" name="optionStart" value="1" checked> <label for="startButton1" id="labelStartButton1" onclick="document.getElementById('infoDemo').style.display='block';document.getElementById('infoDemoAdvanced').style.display='none';document.getElementById('getStarted').style.display='block';document.getElementById('controls').style.display='none'">Getting started</label> <input type="radio" id="startButton2" name="optionStart" value="2" > <label for="startButton2" id="labelStartButton2" onclick="document.getElementById('infoDemo').style.display='none';document.getElementById('infoDemoAdvanced').style.display='block';document.getElementById('getStarted').style.display='none';document.getElementById('controls').style.display='block'">Advanced options</label> </div> </div> <div id="getStarted" class="l-body" style="display:block"> <div id="getStarted1" style="display:block"> <div class="row" > <div id="zoomCrea1" class="column" style="background-color: rgba(0,0,0,0.1);border: solid rgba(0, 0, 0, .2);flex:1; " onclick="img.src='public/zoomCrea1.png'"> <img style="width:100%" src="public/zoomCrea1.png"> <p class="titleGetStarted"> Zoomed in </p> </div> <div id="multi1"class="column" style="background-color: rgba(0,0,0,0.1);border: solid rgba(0, 0, 0, .2);flex:1;" onclick="img.src='public/init1.png'"> <img style="width:100%" src="public/init1.png"> <p class="titleGetStarted"> Multi creature</p> </div> <div id="mazeCrea1" class="column" style="background-color: rgba(0,0,0,0.1);border: solid rgba(0, 0, 0, .2);flex:1;" onclick="img.src='public/mazeCrea1.png'"> <img style="width:100%" src="public/mazeCrea1.png"> <p style="margin-top:0px;margin-bottom:0px;text-align:center"> Maze </p> </div> </div> </div> <div id="getStarted2" style="display:none"> <div class="row" > <div id="zoomCrea2" class="column" style="background-color: rgba(0,0,0,0.1);border: solid rgba(0, 0, 0, .2);flex:1;" onclick="img.src='public/zoomCrea2.png'"> <img style="width:100%" src="public/zoomCrea2.png"> <p class="titleGetStarted"> Zoomed in</p> </div> <div id="multi2"class="column" style="background-color: rgba(0,0,0,0.1);border: solid rgba(0, 0, 0, .2);flex:1;" onclick="img.src='public/init2.png'"> <img style="width:100%" src="public/init2.png"> <p class="titleGetStarted"> Multi creature</p> </div> <div id="mazeCrea2" class="column" style="background-color: rgba(0,0,0,0.1);border: solid rgba(0, 0, 0, .2);flex:1;" onclick="img.src='public/mazeCrea2.png'"> <img style="width:100%" src="public/mazeCrea2.png"> <p class="titleGetStarted"> Maze </p> </div> </div> </div> <div id="getStarted3" style="display:none"> <div class="row" > <div id="zoomCrea3" class="column" style="background-color: rgba(0,0,0,0.1);border: solid rgba(0, 0, 0, .2);flex:1; " onclick="img.src='public/zoomCrea3.png'"> <img style="width:100%" src="public/zoomCrea3.png"> <p class="titleGetStarted"> Zoomed in</p> </div> <div id="multi3"class="column" style="background-color: rgba(0,0,0,0.1);border: solid rgba(0, 0, 0, .2);flex:1;" onclick="img.src='public/init3.png'"> <img style="width:100%" src="public/init3.png"> <p class="titleGetStarted"> Multi creature</p> </div> <div id="mazeCrea3"class="column" style="background-color: rgba(0,0,0,0.1);border: solid rgba(0, 0, 0, .2);flex:1;" onclick="img.src='public/mazeCrea3.png'"> <img style="width:100%" src="public/mazeCrea3.png"> <p style="margin-top:0px;margin-bottom:0px;text-align:center"> Maze </p> </div> </div> </div> </div> <div id="controls" class="l-body" style="display:none"> <div class='row'> <div class="column"> <input type="range" min="0.1" max="1" step="0.1" value="0.5" class="slider" id="rangeRadiusWall" > <p> Brush size: <span id="valueRadiusWall"></span></p> <input type="range" min="6" max="20" step="0.1" value="8.5" class="slider" id="rangeRadius" style="display:none"> <p style="display:none"> Radius of kernels (size of creature): <span id="valueRadius"></span></p> </div> <div class="column"> <input type="range" min="2" max="8" step="1" value="2" class="slider" id="rangeZoom" > <p> Zoom : <span id="valueZoom"></span></p> </div> <div class="column"> <input type="file" id="imageLoader" style="" name="imageLoader"/> </div> </div> <div class="radio-toolbar" id="optionDiv"> <input type="radio" id="radioErase" name='option' value="erase"> <label for="radioErase">Eraser</label> <input type="radio" id="radioCircle" name='option' value="circle" checked> <label for="radioCircle">Wall Dot</label> <input type="radio" id="radioCreature" name='option' value="creature" > <label for="radioCreature">Creature</label> <input type="radio" id="radioZoom" name='option' value="zoom" > <label for="radioZoom">Zoom</label> <input type="radio" id="radioAttract" name='option' value="attract" > <label for="radioAttract">Attractor</label> </div> </div> <script type="text/javascript" src="public/utils.js"></script> <script type="text/javascript" src="public/gl.js"></script> <h2 id='Videos'>Videos</h2> <div class="l-page"> <div class="row"> <div class="column"> <video id="robust" width="95%" controls autoplay loop muted="" class="videoShow"> <source src="public/resultObstacle.mp4" type="video/mp4"> </video> </div> <div class="column"> <video id="sensorimotor_agents" width="95%" controls autoplay loop muted="" class="videoShow"> <source src="public/resultObstacle8.mp4" type="video/mp4"> </video> </div> <div class="column"> <video id="robust" width="95%" controls autoplay loop muted="" class="videoShow"> <source src="public/resultObstacle9.mp4" type="video/mp4"> </video> </div> <div class="column"> <video id="robust" width="95%" controls autoplay loop muted="" class="videoShow"> <source src="public/robust2.mp4" type="video/mp4"> </video> </div> </div> <p style="font-size: 19px;color:#000000;text-align:center;" > Video S1: Sensorimotor agents<br> <span style="font-size: 14px;color:#A0A0A0;"> Different agents (yellow) emerging from rules obtained by the IMGEP. The agents display sensorimotor capabilities: they are robust and react to perturbations by the obstacles (blue). The righmost video shows the system with a different colormap (fixed obstacle channel in black) to highlight the differences in activity in the agent as a response to perturbation. </span> </p> </div> <div class="l-page"> <video id="random_trials" width="95%" controls muted="" class="videoShow"> <source src="public/random.mp4" type="video/mp4"> </video> <p style="font-size: 19px;color:#000000;text-align:center;" > Video S2: Random search<br> <span style="font-size: 14px;color:#A0A0A0;"> Each 100 squares are random parameters trials (each 1 channel and 10 rules so \(\sim \) 130 parameters for all the rules of a square). We observe that a lot of random search trials lead to death or explosion of the mass. Very little lead to stable spatially localized pattern and even less to moving ones. </span> </p> <div class="container_row l-page"> <div class="column"> <video id="robust" width="99%" controls loop muted="" class="videoShow"> <source src="public/orbiumSolo.mp4" type="video/mp4"> </video> </div> <div class="column"> <video id="orbium_collision" width="99%" controls loop muted="" class="videoShow"> <source src="public/orbiumCollision.mp4" type="video/mp4"> </video> </div> <div class="column"> <video id="robust" width="99%" controls loop muted="" class="videoShow"> <source src="public/orbiumCollision2.mp4" type="video/mp4"> </video> </div> </div> <p style="font-size: 19px;color:#000000;text-align:center;" > Video S3: Orbium, moving agent from the original lenia papers, fragile to external perturbations<br> <span style="font-size: 14px;color:#A0A0A0;"> Left: Orbium: the equivalent of the glider in Lenia (from the original lenia paper), an example of moving agent. Middle and right videos: collision between several orbium leading to death/explosion. This shows the fragility of the orbium to external perturbations. </span> </p> <div class="l-screen-inset"> <div id="orbicol"> <div class="container_row"> <div class="column"></div> <div class="column"> <video id="robust" width="95%" controls loop muted="" class="videoShow"> <source src="public/orbium.mp4" type="video/mp4"> </video> </div> <div class="column"> <video id="orbium_obstacles" width="95%" controls loop muted="" class="videoShow"> <source src="public/orbiumWallb.mp4" type="video/mp4"> </video> </div> <div class="column"></div> </div> <p style="font-size: 19px;color:#000000;text-align:center;" > Video S4: Orbium perturbed by obstacles. <br> <span style="font-size: 14px;color:#A0A0A0;"> Orbium, equivalent of the glider in Lenia (from the original lenia paper), dies from perturbations by obstacles. </span> </p> </div> </div> <div class="l-screen-inset"> <div id="overview_agents" class="row"> <div class="column"> <video id="robust" width="95%" controls loop muted="" class="videoShow"> <source src="public/random_overwiew_contour.mp4" type="video/mp4"> </video> <p style="text-align:center;"> Random search</p> </div> <div class="column"> <video id="sensorimotor_agents" width="95%" controls loop muted="" class="videoShow"> <source src="public/imgep_overview_contour.mp4" type="video/mp4"> </video> <p style="text-align:center;"> IMGEP search </p> </div> <div class="column"> <video id="robust" width="95%" controls loop muted="" class="videoShow"> <source src="public/handmade_overwiew_contour.mp4" type="video/mp4"> </video> <p style="text-align:center;"> Handmade search </p> </div> </div> <p style="font-size: 19px;color:#000000;text-align:center;" > Video S5: Agents obtained by each method<br> <span style="font-size: 14px;color:#A0A0A0;"> 100 Patterns passing our agency tests obtained by each method (from left to right): random search,IMGEP, handmade search (from Lenia original papers). A lot of IMGEP obtained agents are moving agents with high speed while a lot of agents obtained by random search are static. </span> </p> </div> <h3 id='Videos'>Generalization</h3> <div class="l-screen-inset"> <div id="moveobs_overview" class="row"> <div class="column"> <video id="robust" width="95%" controls loop muted="" class="videoShow"> <source src="public/random_movobsb.mp4" type="video/mp4"> </video> <p style="text-align:center;"> Random search</p> </div> <div class="column"> <video id="sensorimotor_agents" width="95%" controls loop muted="" class="videoShow"> <source src="public/imgep_movobsb.mp4" type="video/mp4"> </video> <p style="text-align:center;"> IMGEP search </p> </div> <div class="column"> <video id="robust" width="95%" controls loop muted="" class="videoShow"> <source src="public/handmade_movobsb.mp4" type="video/mp4"> </video> <p style="text-align:center;"> Handmade search </p> </div> </div> <p style="font-size: 19px;color:#000000;text-align:center;" > Video S6: Moving obstacle test on agents obtained by each method<br> <span style="font-size: 14px;color:#A0A0A0;"> 100 Agents obtained by random search,IMGEP, handmade search (from Lenia original papers). We observe that the proportion of agents with robustness to moving obstacles is much higher in the agents obtained by IMGEP than the ones obtained by random search and handmade search. </span> </p> </div> <h4> Quantitative</h4> <div class="l-screen-inset"> <div class="row "> <div class="column"> <p style="font-size: 12px;color:#000000;text-align:center;">Seed 9 IMGEP step 142</p> </div> <div class="column"> <video id="kernels" width="100%" controls loop muted="" class="videoShow" > <source src="public/generalization_seed9_142/seed9_crea00142_obstacle_number_var_30.00.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.1<br> <span style="font-size: 10px;color:#A0A0A0;"> Obstacle number : 30 </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed9_142/seed9_crea00142_obstacle_radius_var_7.00.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.2<br> <span style="font-size: 10px;color:#A0A0A0;"> Obstacle radius 7 </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed9_142/seed9_crea00142_obstacle_speed_var_3.00.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.3<br> <span style="font-size: 10px;color:#A0A0A0;"> Obstacle speed 3 </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed9_142/seed9_crea00142_update_mask_rate_var_0.20.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.4<br> <span style="font-size: 10px;color:#A0A0A0;"> Update mask rate 0.2 </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed9_142/seed9_crea00142_update_noise_rate_var_0.80.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.5<br> <span style="font-size: 10px;color:#A0A0A0;"> Update noise rate 0.8 </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed9_142/seed9_crea00142_update_noise_std_var_0.60.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.6<br> <span style="font-size: 10px;color:#A0A0A0;"> Update noise std 0.6 </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed9_142/seed9_crea00142_init_noise_rate_var_1.00.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.7<br> <span style="font-size: 10px;color:#A0A0A0;"> Init noise rate 1 </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed9_142/seed9_crea00142_init_noise_std_var_4.50.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.8<br> <span style="font-size: 10px;color:#A0A0A0;"> Init noise std 4.5 </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed9_142/seed9_crea00142_scaling_var_2.15.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.9<br> <span style="font-size: 10px;color:#A0A0A0;"> Rescaling 2.15 </span> </p> </div> </div> <div class="row "> <div class="column"> <p style="font-size: 12px;color:#000000;text-align:center;">Seed 5 IMGEP step 120</p> </div> <div class="column"> <video id="kernels" width="100%" controls="" loop muted="" class="videoShow" > <source src="public/generalization_seed5_120/seed5_crea00120_obstacle_number_var_30.00.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.11<br> <span style="font-size: 10px;color:#A0A0A0;"> Obstacle number : 30 </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed5_120/seed5_crea00120_obstacle_radius_var_7.00.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.12<br> <span style="font-size: 10px;color:#A0A0A0;"> Obstacle Radius 7. </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed5_120/seed5_crea00120_obstacle_speed_var_3.00.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.13<br> <span style="font-size: 10px;color:#A0A0A0;"> Obstacle speed 3 </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed5_120/seed5_crea00120_update_mask_rate_var_0.20.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.14<br> <span style="font-size: 10px;color:#A0A0A0;"> Update mask rate 0.2 </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed5_120/seed5_crea00120_update_noise_rate_var_0.80.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.15<br> <span style="font-size: 10px;color:#A0A0A0;"> Update noise rate 0.8 </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed5_120/seed5_crea00120_update_noise_std_var_0.60.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.16<br> <span style="font-size: 10px;color:#A0A0A0;"> Update noise std 0.6 </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed5_120/seed5_crea00120_init_noise_rate_var_1.00.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.17<br> <span style="font-size: 10px;color:#A0A0A0;"> Init noise rate 1. </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed5_120/seed5_crea00120_init_noise_std_var_4.50.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.18<br> <span style="font-size: 10px;color:#A0A0A0;"> Init noise std 4.5 </span> </p> </div> <div class="column"> <video id="robust" width="100%" controls loop muted="" class="videoShow"> <source src="public/generalization_seed5_120/seed5_crea00120_scaling_var_2.15.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S7.19<br> <span style="font-size: 10px;color:#A0A0A0;"> Rescaling 2.15 </span> </p> </div> </div> <p style="font-size: 19px;color:#000000;text-align:center;" > Video S7 : Illustration of the quantitative tests performed<br> <span style="font-size: 14px;color:#A0A0A0;"> Videos of quantitative tests for 2 moving agents obtained by IMGEP. We display only a subset of the value tested for every quantitative test. </span> </p> </div> <h4> Qualitative </h4> <div class="l-screen-inset"> <div class="container_row"> <div class="column"> <video id="robust" width="95%" controls="" loop muted="" class=" videoShow"> <source src="public/custom_obstacles.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S8: Out of distribution obstacles: Different shapes<br> <span style="font-size: 10px;color:#A0A0A0;"> Test of a moving agent obtained by IMGEP on obstacles that were not seen during training. </span> </p> </div> <div class="column"> <video id="robust" width="95%" controls="" loop muted="" class="videoShow"> <source src="public/solveMaze.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S9: Out of distribution obstacles: maze.<br> <span style="font-size: 10px;color:#A0A0A0;"> Test of a moving agent to maze like obstacles. </span> </p> </div> <div class="column"> <video id="bullets" width="95%" controls loop muted="" class="videoShow "> <source src="public/harderEnv.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S10: Out of distribution obstacles: Bullet like obstacles<br> <span style="font-size: 10px;color:#A0A0A0;"> Test of a moving agent to bullet like environment: fast small moving obstacles. </span> </p> </div> </div> </div> <div class="l-screen-inset"> <div class="container_row"> <div class="column"> <video id="individuality" width="95%" controls="" loop muted="" class="videoShow"> <source src="public/demo.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S11: Individuality preservation<br> <span style="font-size: 10px;color:#A0A0A0;"> Example of moving agents obtained by IMGEP colliding while keeping their individuality, they don't merge or collapse from the collision. </span> </p> </div> <div class="column"> <video id="reproduction" width="95%" controls="" loop muted="" class="videoShow"> <source src="public/repro.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S12: Reproduction<br> <span style="font-size: 10px;color:#A0A0A0;"> For some moving agents, under specific conditions, the collision of 2 agents can lead to the self-organization of a 3rd agent. (each with its own individuality) </span> </p> </div> <div class="column"> <video id="attraction" width="95%" controls="" loop muted="" class="videoShow"> <source src="public/stick.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S13: Attraction<br> <span style="font-size: 10px;color:#A0A0A0;"> Example of moving agents attracting each other while still maintaining their own individuality. </span> </p> <!-- <video id="robust" width="95%" controls loop muted="" class="videoShow"> <source src="public/stickFar.mp4" type="video/mp4"> </video> <p style="font-size: 13px;color:#A0A0A0; "> If they are too far from each other no attraction. </p> </div>--> </div> </div> </div> <div class="l-screen-inset"> <div class="container_row "> <div class="column"> <video id="asynchro" width="95%" controls="" loop muted="" class="videoShow"> <source src="public/asynchro.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S14: Asynchronous update<br> <span style="font-size: 10px;color:#A0A0A0;"> Testing a moving agent with asynchronous updates. Each cell is updated with a certain probability at each step leading to cells being asynchronously updated. </span> </p> </div> <div class="column"> <video id="scaling" width="95%" controls="" loop muted="" class="videoShow"> <source src="public/smallerDie.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S15: Scaling the agents down<br> <span style="font-size: 10px;color:#A0A0A0;"> The moving agents size is reduced. The scaled down agents still seem to behave similarly (same shape and have sensorimotor capabilities) to the normal size one while being composed of less cells. </span> </p> </div> <div class="column"> <video id="damage_hand" width="95%" controls="" loop muted="" class="videoShow"> <source src="public/damageHand.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S16: Morphological computation<br> <span style="font-size: 10px;color:#A0A0A0;"> We pause the simulation and remove some cells of a moving agent. As a response to this alteration of the structure, the moving agent changes direction, regrow itself and moves away. This video isolates the fact that the macro agent senses perturbations of its structure and respond to it by a morphological growth. </span> </p> </div> </div> </div> <div class="l-screen-inset"> <div class="container_row"> <div class="column"></div> <div class="column"> <video id="attraction_external" width="95%" controls loop muted="" class="videoShow"> <source src="public/attractLive.mp4" type="video/mp4"> </video> <p style="font-size: 12px;color:#000000;text-align:center;" > Video S17: External control.<br> <span style="font-size: 10px;color:#A0A0A0;"> We introduce an attractive element in another channel (in Cyan). We learned the rule that control the way this external element channel influences the learnable channel(Yellow) and display the resulting behavior here. The moving agent is effectively attracted to this introduce component. By controlling the external element we can control live the direction of the moving agent. </span> </p> </div> <div class="column"></div> </div> </div> <div class="l-screen-inset"> <div class="container_row"> <div class="column" > <video id="model-editing-level-1-video-1" width="95%" controls="" muted="" class="videoShow "> <source src="public/initCircleGrad.mp4" type="video/mp4"> </video> </div> <div class="column"> <video id="model-editing-level-1-video-1" width="95%" controls="" muted="" class="videoShow "> <source src="public/initCircle.mp4" type="video/mp4"> </video> </div> <div class="column"> <video id="init_death" width="95%" controls="" muted="" class="videoShow "> <source src="public/initDie.mp4" type="video/mp4"> </video> </div> </div> <p style="font-size: 19px;color:#000000;text-align:center;" > Video S18: Robustness to initialization<br> <span style="font-size: 14px;color:#A0A0A0;"> Testing the robustness of the learned rule to emerge an agent from different initial patterns. We replace the learned initial pattern by : left a disk with a gradient; middle a large disk (much larger than an agent); right right top a disk with gradient of another size, right bottom a disk without gradient. Some initialization lead to the robust emergence of one or several agents while some lead to the collapse of the pattern. </span> </p> </div> <h3> additional </h3> <div class="l-page"> <div class="container_row"> <div class="column"> <video id="non_moving" width="95%" controls loop muted="" class="videoShow"> <source src="public/no_moving_seed0_crea00147.mp4" type="video/mp4"> </video> </div> <div class="column"> <video id="robust" width="95%" controls loop muted="" class="videoShow"> <source src="public/no_moving_seed1_crea00101.mp4" type="video/mp4"> </video> </div> </div> <p style="font-size: 19px;color:#000000;text-align:center;" > Video S19: Examples of agents considered non moving by our moving test<br> </p> </div> <div class="l-page"> <video id="robust" width="100%" controls="" autoplay loop muted="" class="videoShow"> <source src="public/unstable.mp4" type="video/mp4"> </video> <p style="font-size: 19px;color:#000000;text-align:center;" > Video S20: Unstability<br> <span style="font-size: 14px;color:#A0A0A0;"> Example of partial robustness of a moving agent. The agent is able to deal with some perturbation by obstacle but some other might break the equilibrium (leading to explosion) </span> </p> </div> <div class="l-page"> <video id="kernelsVis" width=85% controls="" muted="" class="videoShow l-page" > <source src="public/kernels.mp4" type="video/mp4"> </video> <p style="font-size: 19px;color:#000000;text-align:center;" > Video S21: Visualization of each rule<br> <span style="font-size: 14px;color:#A0A0A0;"> For the visualization of the role of each rule, we display the growth (in the magma colormap) of each of the 10 rules in the exterior videos. The inner top video shows the normal simulation. The inner bottom video displays the weighted sum of the growth of all the rules in grey level (1 being white and 0 being black), showing what is added to the current state of the learnable channel. </span> </p> </div> </dt-article> <script type="text/javascript"> "use strict"; function changeImgIMGEP(img) { var schemeIMGEP=document.getElementById("schemeIMGEP"); schemeIMGEP.src=img; } function changeVideo(className,nb) { var vids=document.getElementsByClassName(className); for(var i=0;i<vids.length;i++){ vids[i].style.display="none"; } console.log(className+nb); var vid=document.getElementById(className+nb); console.log(vid); vid.style.display="block"; vid.currentTime = '0'; } </script> <style> button { display: inline-block; background-color: #7b38d8; border-radius: 10px; border: 4px double #cccccc; color: #eeeeee; text-align: center; font-size: 15px; padding: 10px; width: 80px; -webkit-transition: all 0.5s; -moz-transition: all 0.5s; -o-transition: all 0.5s; transition: all 0.5s; cursor: pointer; margin: 5px; } button:hover { background-color: lightgreen; } .radio-toolbar input[type="radio"] { opacity: 0; position: fixed; width: 0; } .titleGetStarted{ margin-top:0px; margin-bottom:0px; text-align:center; } .radio-toolbar label { display: inline-block; background-color: #cbc; border-radius: 10px; border: 4px double #cccccc; color: #eeeeee; text-align: center; font-size: 15px; padding: 10px; width: 80px; -webkit-transition: all 0.5s; -moz-transition: all 0.5s; -o-transition: all 0.5s; transition: all 0.5s; cursor: pointer; margin: 5px; } .radio-toolbar input[type="radio"]:checked + label { background-color:green; border-color: #4c4; } .radio-toolbar input[type="radio"]:focus + label { border: 2px dashed #444; } .radio-toolbar label:hover { background-color: lightgreen; } .figcaption{ font-size: 22px; color:#A0A0A0; } d-content li{ font-size: 14px; color:#A0A0A0; } d-content { clear: both; float: right; line-height:1.4; padding-top: 1em; border-left: 1px solid rgba(0, 0, 0, 0.1); margin-top: 0; margin-left: 10px; margin-right: calc((100vw - 1500px) / 2 + 168px); width: calc((1500px - 648px) / 2 - 24px); } d-content h5{ margin-block-start: 1.5em; margin-block-end: 0em; } d-content ul{ margin-block-start: 0.5em; margin-block-end: 0.5em; } d-content h2{ margin-top:10px; } @media (max-width: 1300px) d-content { display: none; } .icon { width: 30px; height: 30px; background: steelblue; fill: white; border-radius: 20px; padding: 5px; margin: 2px; cursor: pointer; } def{ clear: both; float: left; margin-top: 0; margin-left: 24px; width: calc(50% - 1384px / 2); padding: 0 0 0 24px; border-radius: 20px 20px 20px 20px; background-color: rgba(0,0,0,0.05); color: rgba(0,0,0,0.5); font-size:14px; } .container_row { display: grid; grid-template-columns: repeat(auto-fit, minmax(100px, 1fr)); grid-gap: 5px; } </style>