We have all seen how ChatGPT can instantly summarize a document or write a poem. But if you look at the world of robotics, the progress feels strangely sluggish. While your phone can do things that felt like science fiction a few years ago, the robots in our factories and warehouses still struggle with tasks a toddler could master, like picking up an object they have never seen before or navigating a crowded hallway without getting stuck.
The robotics industry is searching for a breakthrough that mimics the sudden, widespread adoption of language-based AI. Currently, experts are trying to solve a fundamental mismatch: language models like ChatGPT were trained on the entire internet, giving them a vast library of human knowledge. Robots, however, lack an equivalent library of physical experiences. They cannot simply download the internet to learn how to operate in the real world. They need to understand physics, gravity, and the unpredictable movement of objects, which requires a completely different kind of data.
Why robots are still clumsy
To teach an AI to write, you show it billions of sentences so it can learn patterns in how words relate to each other. When you ask it a question, it predicts the most likely next word. To teach a robot to act, you need to connect an AI model—the digital engine that processes information—to physical sensors like cameras and touch-sensitive motors. This is called Physical AI.
The challenge is that the real world is messy. In a simulation, you can create a perfect digital version of a room, but a robot often fails when it steps into a room with different lighting, weird shadows, or misplaced furniture. Because we do not have a giant dataset of every possible physical interaction, companies are now trying to create this data artificially. They use high-powered computers to run millions of simulations, effectively forcing robots to practice thousands of times in a virtual world before they ever try to lift a real box.
The hurdle here is not just intelligence; it is trust and safety. Because these robots rely on AI to interpret what they see, they are vulnerable to new types of digital attacks. A researcher might place a specific sticker on a floor that causes a robot to get confused, or use a cleverly crafted digital signal to make a robot ignore its safety stops. As we bring more robots into our workplaces and homes, we are moving from a world where we only worried about hardware breaking to a world where we must worry about the software being tricked.
The goal is to build robots that can handle the unexpected. We are currently in a "training phase" for the entire field of robotics, trying to move from machines that only follow rigid, pre-programmed instructions to machines that can actually "understand" the physical space around them. Until they can navigate that uncertainty as naturally as ChatGPT handles language, they will remain locked in our factories and labs rather than walking around our living rooms.
Liked this one? The next lands at breakfast.
Every story in tomorrow's AI news, rebuilt in plain English — five minutes, sources linked, free forever.
By joining you agree to receive Article's daily newsletter — unsubscribe in one click. Privacy