[{"data":1,"prerenderedAt":128},["ShallowReactive",2],{"news-\u002Fnews\u002Ferste-schritte-simulation-teleoperation":3,"news-en-\u002Fnews\u002Ferste-schritte-simulation-teleoperation":74},{"id":4,"title":5,"author":6,"body":7,"date":55,"description":56,"draft":57,"extension":58,"image":59,"links":60,"meta":61,"name":62,"navigation":63,"path":64,"seo":65,"stem":66,"tags":67,"__hash__":73},"news\u002Fnews\u002Ferste-schritte-simulation-teleoperation.md","Erste Schritte: vom ferngesteuerten zum KI-gesteuerten Roboter","Symbiosika Team",{"type":8,"value":9,"toc":47},"minimark",[10,14,18,23,26,30,33,37,40,44],[11,12,5],"h1",{"id":13},"erste-schritte-vom-ferngesteuerten-zum-ki-gesteuerten-roboter",[15,16,17],"p",{},"Seit Jahresbeginn arbeiten wir an den Grundlagen. Bevor ein Roboter selbstständig arbeitet, muss man ihn erst einmal sicher bewegen, beobachten und messen können.",[19,20,22],"h2",{"id":21},"teleoperation-in-der-simulation","Teleoperation in der Simulation",[15,24,25],{},"Wir haben eine Umgebung in NVIDIA Isaac Sim aufgebaut, in der wir Roboter per Fernsteuerung durch virtuelle Räume führen. Jede Bewegung wird dabei aufgezeichnet. So entstehen die ersten Datensätze, aus denen Roboter später lernen.",[19,27,29],{"id":28},"ein-roboter-den-ein-ki-agent-steuert","Ein Roboter, den ein KI-Agent steuert",[15,31,32],{},"Parallel haben wir eine Kommandozeilen-Schnittstelle gebaut, über die ein KI-Agent einen realen Roboter steuert. Statt Gelenkwinkel einzugeben, formuliert man das Ziel, und der Agent übersetzt es in Bewegungen.",[19,34,36],{"id":35},"navigation-und-objekterkennung","Navigation und Objekterkennung",[15,38,39],{},"In einer simulierten Küche fährt unser Roboter inzwischen selbstständig zu benannten Bereichen wie „Küchentisch“ oder „Regal“. Dort erkennt er Objekte mit einem Kameramodell, bestimmt über die Tiefenkamera ihre Position im Raum und fährt gezielt heran.",[19,41,43],{"id":42},"was-wir-gelernt-haben","Was wir gelernt haben",[15,45,46],{},"Simulation spart enorm viel Zeit, aber nur, wenn sie realistisch genug ist. Spiegelnde und durchsichtige Gegenstände sind zum Beispiel für Tiefenkameras schwierig. Solche Grenzen früh zu kennen, ist mindestens so wertvoll wie jede gelungene Demo.",{"title":48,"searchDepth":49,"depth":49,"links":50},"",2,[51,52,53,54],{"id":21,"depth":49,"text":22},{"id":28,"depth":49,"text":29},{"id":35,"depth":49,"text":36},{"id":42,"depth":49,"text":43},"2026-03-17T00:00:00.000Z","In den ersten Monaten haben wir eine Simulationsumgebung aufgebaut, Roboter ferngesteuert und erstmals per KI-Agent bewegt.",false,"md","\u002Fimages\u002Flanding\u002Frobot-portrait.webp",null,{},"erste-schritte-simulation-teleoperation",true,"\u002Fnews\u002Ferste-schritte-simulation-teleoperation",{"title":5,"description":56},"news\u002Ferste-schritte-simulation-teleoperation",[68,69,70,71,72],"Robotik","Simulation","Isaac Sim","Navigation","Objekterkennung","7Q21xH1ONohOXvlRs-IMSkT6Mb-KtY14olj_y5v0HHE",{"id":75,"title":76,"author":6,"body":77,"date":55,"description":119,"draft":57,"extension":58,"image":59,"links":60,"meta":120,"name":62,"navigation":63,"path":121,"seo":122,"stem":123,"tags":124,"__hash__":127},"newsEn\u002Fnews-en\u002Ferste-schritte-simulation-teleoperation.md","First steps: from a remote-controlled to an AI-controlled robot",{"type":8,"value":78,"toc":113},[79,82,85,89,92,96,99,103,106,110],[11,80,76],{"id":81},"first-steps-from-a-remote-controlled-to-an-ai-controlled-robot",[15,83,84],{},"Since the start of the year we have been working on the foundations. Before a robot can work on its own, you first need to be able to move it safely, observe it and measure it.",[19,86,88],{"id":87},"teleoperation-in-simulation","Teleoperation in simulation",[15,90,91],{},"We have built an environment in NVIDIA Isaac Sim where we guide robots through virtual rooms by remote control. Every movement is recorded. These are the first datasets robots will later learn from.",[19,93,95],{"id":94},"a-robot-controlled-by-an-ai-agent","A robot controlled by an AI agent",[15,97,98],{},"In parallel, we have built a command-line interface through which an AI agent controls a real robot. Instead of entering joint angles, you state the goal, and the agent translates it into movements.",[19,100,102],{"id":101},"navigation-and-object-detection","Navigation and object detection",[15,104,105],{},"In a simulated kitchen, our robot now drives on its own to named areas such as \"kitchen table\" or \"shelf\". There it detects objects with a camera model, uses the depth camera to determine their position in the room and approaches them.",[19,107,109],{"id":108},"what-we-have-learned","What we have learned",[15,111,112],{},"Simulation saves an enormous amount of time, but only if it is realistic enough. Reflective and transparent objects, for example, are hard for depth cameras. Knowing such limits early is at least as valuable as any successful demo.",{"title":48,"searchDepth":49,"depth":49,"links":114},[115,116,117,118],{"id":87,"depth":49,"text":88},{"id":94,"depth":49,"text":95},{"id":101,"depth":49,"text":102},{"id":108,"depth":49,"text":109},"We have set up a simulation environment, teleoperated robots and moved one with an AI agent for the first time.",{},"\u002Fnews-en\u002Ferste-schritte-simulation-teleoperation",{"title":76,"description":119},"news-en\u002Ferste-schritte-simulation-teleoperation",[125,69,70,71,126],"Robotics","Object detection","5s-dcobJda6EWF3orKXMvzsaGk7DgnyhbrEt11tOccg",1792586944992]