That little apricot gave me more trouble than I expected 😂 Spent some time on @axisrobotics trying to guide a Wheeled Franka robot through a simple pick and place task. One thing I kept noticing was how different the scene looked from each camera angle. From one view, the gripper seemed perfectly aligned with the apricot. Switch to another, and suddenly I wasn't so sure anymore. It made me realize something interesting about Physical AI. Knowing where an object is doesn't necessarily mean knowing how to reach it. The robot needs the right approach, the right position, and enough spatial awareness to interact with the object without missing it. That's why I find browser-based teleoperation so interesting. Human control can provide examples of how these small physical decisions are made. Still working on my robot driving skills, though. The apricot is probably judging me. 🍑😭 #AxisRobotics# #PhysicalAI#

I've been thinking about something lately 😭 When we talk about Physical AI, we usually focus on the moment a robot successfully completes a task. But there's another story behind that moment. Someone had to design the task. Someone had to collect the demonstrations. And those recordings had to become usable training data. That's what makes @axisrobotics interesting to me. Axis is building infrastructure that connects task generation, human contributed simulation data, and data processing into a broader robot learning workflow. The robot's final movement might be the most visible part. But the work that makes learning possible happens long before that. A successful robot action is what we see. The data infrastructure behind it is what we rarely notice. And honestly, I think that hidden part deserves more attention. 🧡 #AxisRobotics# #PhysicalAI#

WAIT… WHAT IF THE OBJECT ISN'T THERE WHEN I REACH FOR IT? 😭 I opened this @axisrobotics conveyor task thinking: okay, find the right item, pick it up, drop it in the basket. sounds easy enough. Then I looked at the setup. Two robot arms. A basket. Several objects. Two camera views. And a conveyor. One arm is holding the basket while the other reaches for the purple container. And suddenly I had a question: What if the object moves between the moment I see it and the moment I try to grab it? 😭 The camera shows a position at one moment. But the gripper reaches that position later. If the object is moving, the robot can't simply rely on where it was a moment ago. It has to account for the timing of the interaction. OBSERVE ↓ APPROACH ↓ GRASP ↓ PLACE Sounds like four simple steps. But each step takes time. And the physical world doesn't necessarily wait. That's what made this @axisrobotics task interesting to me. Through browser based teleoperation, contributors perform simulated manipulation tasks and create interaction trajectories that can be processed for robot learning. It's not just about moving an arm from point A to point B. It's about performing actions in a physical environment where timing can matter. Seeing the right object is one challenge. Reaching it at the right moment is another. And honestly, I didn't expect a little purple container to make me think this much about Physical AI. 😭 #AxisRobotics# #PhysicalAI#