[{"data":1,"prerenderedAt":143},["ShallowReactive",2],{"news-\u002Fnews\u002Froboter-planung":3,"news-en-\u002Fnews\u002Froboter-planung":82},{"id":4,"title":5,"author":6,"body":7,"date":64,"description":65,"draft":66,"extension":67,"image":68,"links":69,"meta":70,"name":71,"navigation":72,"path":73,"seo":74,"stem":75,"tags":76,"__hash__":81},"news\u002Fnews\u002Froboter-planung.md","Planen statt programmieren: wie unser Roboter Aufgaben selbst zerlegt","Symbiosika Team",{"type":8,"value":9,"toc":57},"minimark",[10,14,18,23,40,43,47,50,54],[11,12,5],"h1",{"id":13},"planen-statt-programmieren-wie-unser-roboter-aufgaben-selbst-zerlegt",[15,16,17],"p",{},"In den letzten Monaten haben wir die einzelnen Bausteine zusammengeführt: Navigation, Objekterkennung, die Plattform zur Verwaltung und das Flywheel für Trainingsdaten. Das Ergebnis ist eine Architektur, die die Steuerung unseres Roboters grundlegend vereinfacht.",[19,20,22],"h2",{"id":21},"zwei-ebenen-des-denkens","Zwei Ebenen des Denkens",[24,25,26,34],"ul",{},[27,28,29,33],"li",{},[30,31,32],"strong",{},"Der Planer",": Ein Sprach- und Bildmodell versteht den Auftrag und die Umgebung und entscheidet, welche Schritte in welcher Reihenfolge nötig sind.",[27,35,36,39],{},[30,37,38],{},"Die Fähigkeiten-Bibliothek",": Jede Fähigkeit, ob Fahren, Greifen, Ablegen oder Melden, ist einzeln getestet und kann beliebig kombiniert werden.",[15,41,42],{},"Der Planer ruft diese Fähigkeiten direkt auf. Eine neue Aufgabe heißt damit meist nur: eine neue Kombination bestehender Fähigkeiten, nicht neuer Code.",[19,44,46],{"id":45},"offen-angebunden","Offen angebunden",[15,48,49],{},"Der Roboter ist über eine einheitliche Schnittstelle steuerbar: über eine Web-API, über die Kommandozeile oder über das Model Context Protocol (MCP) direkt aus einem KI-Agenten heraus. Damit lässt er sich leicht in bestehende Systeme eines Hauses einbinden.",[19,51,53],{"id":52},"erprobt-auf-mehreren-plattformen","Erprobt auf mehreren Plattformen",[15,55,56],{},"In der Simulation läuft diese Architektur bereits auf einem humanoiden Roboter und auf einem mobilen Manipulator. Die nächsten Schritte: mehr Fähigkeiten, mehr Szenarien aus Hotel und Pflege und der Weg aus der Simulation in reale Testumgebungen.",{"title":58,"searchDepth":59,"depth":59,"links":60},"",2,[61,62,63],{"id":21,"depth":59,"text":22},{"id":45,"depth":59,"text":46},{"id":52,"depth":59,"text":53},"2026-10-14T00:00:00.000Z","Ein Sprachmodell plant, eine Bibliothek getesteter Fähigkeiten führt aus. So bringen wir neue Aufgaben auf den Roboter, ohne jedes Mal neu zu programmieren.",false,"md","\u002Fimages\u002Fkinaut\u002FKINAUT_ONE_Hotel_1.webp",null,{},"roboter-planung",true,"\u002Fnews\u002Froboter-planung",{"title":5,"description":65},"news\u002Froboter-planung",[77,78,79,80],"Robotik","KI","Sprachmodelle","Simulation","8EhtgUfdBv7LIo6dojFM7nXpN0thEPEPHiiQseVK8bs",{"id":83,"title":84,"author":6,"body":85,"date":64,"description":133,"draft":66,"extension":67,"image":68,"links":69,"meta":134,"name":71,"navigation":72,"path":135,"seo":136,"stem":137,"tags":138,"__hash__":142},"newsEn\u002Fnews-en\u002Froboter-planung.md","Planning instead of programming: how our robot breaks down tasks itself",{"type":8,"value":86,"toc":128},[87,90,93,97,111,114,118,121,125],[11,88,84],{"id":89},"planning-instead-of-programming-how-our-robot-breaks-down-tasks-itself",[15,91,92],{},"Over the last few months we have brought the individual building blocks together: navigation, object detection, the management platform and the training-data flywheel. The result is an architecture that fundamentally simplifies how our robot is controlled.",[19,94,96],{"id":95},"two-levels-of-thinking","Two levels of thinking",[24,98,99,105],{},[27,100,101,104],{},[30,102,103],{},"The planner",": A language and vision model understands the instruction and the environment and decides which steps are needed in which order.",[27,106,107,110],{},[30,108,109],{},"The skill library",": Every skill, whether driving, grasping, placing or reporting, is tested on its own and can be combined freely.",[15,112,113],{},"The planner calls these skills directly. A new task usually just means a new combination of existing skills, not new code.",[19,115,117],{"id":116},"openly-connected","Openly connected",[15,119,120],{},"The robot can be controlled through a single interface: a web API, the command line, or the Model Context Protocol (MCP) directly from an AI agent. That makes it easy to integrate into a building's existing systems.",[19,122,124],{"id":123},"tested-on-several-platforms","Tested on several platforms",[15,126,127],{},"In simulation, this architecture already runs on a humanoid robot and on a mobile manipulator. Next steps: more skills, more hotel and care scenarios, and the move from simulation into real test environments.",{"title":58,"searchDepth":59,"depth":59,"links":129},[130,131,132],{"id":95,"depth":59,"text":96},{"id":116,"depth":59,"text":117},{"id":123,"depth":59,"text":124},"A language model plans, a library of tested skills executes. This lets us bring new tasks to the robot without reprogramming it every time.",{},"\u002Fnews-en\u002Froboter-planung",{"title":84,"description":133},"news-en\u002Froboter-planung",[139,140,141,80],"Robotics","AI","Language models","uKk7TM01RJBNXD6ovjUCst86SCP7n03UqbZFnNcxdpg",1792586944992]