I Saw the Future of Robotics in a Virtual Duck That Chases My Cat

WIRED's Will Knight used OpenAI's GPT-6 Sol, Nous Research's Hermes agent harness, Nvidia's Isaac Sim and a DGX Spark to teach a virtual Microduck, a small robot from Pollen Robotics and Hugging Face, to walk without falling and to find his Bengal cat. He says the job took a few hours that once might have taken him weeks.

That is a cute anecdote, but it sits on top of serious lab work. UC Berkeley's Ken Goldberg says OpenAI's Astra built a full 3D simulation from a video clip in about an hour, and Nvidia's Jim Fan and collaborators just open sourced Aspire, an agentic framework that lets a robot discover and use new skills.

We think this is one of the more important shifts in robotics this year, and it is easy to miss because it shows up as a duck chasing a cat. The bottleneck in robot learning has long been the simulation grind: building accurate scenes, tuning parameters, running experiments, watching results, adjusting. What Knight describes is a model doing that loop itself, writing the code, watching video of the robot, and proposing training changes.

If that holds up, the cost of getting a robot to do a new task drops sharply, and the pool of people who can try expands from specialist labs to hobbyists and small manufacturers. That is good news. Goldberg says industrial robotics and automation firms are already interested, and WIRED points to manufacturing and e-commerce as places this could land fast.

But let's be honest about the limits the piece itself admits. The moonwalk attempt ended as a jaunty backward jog. Knight's duck vision system isn't good enough for a real product. Getting robots to work reliably still takes deep engineering. Simulation that looks right is not the same as behavior that is safe in a kitchen, a warehouse, or next to a pet.

That is where I'm watching the "vibe robotics" idea most carefully. Vibe coding gave us a lot of software that worked in a demo and broke in production. Broken software loses data; broken robots can hurt people and property. If model-trained behavior moves into physical machines faster than testing practices do, the failures will be physical. The upside is real. The discipline has to keep pace.

Sim-to-real transfer, training a robot in a virtual environment and then running that behavior on hardware, depends on simulations that are detailed and accurate. Nvidia's Isaac Sim and Newton are widely used tools for that work. The Conference on Robot Learning, where Fan spoke this week, is a key venue for the field.