Traditionally robots are automation tools you pre-program with an exact sequence of motions, like packaging items into a box. But clothes are deformable — the shapes are always different and not in rigid states — so you can't pre-program a sequence of motions to fold cloth. It needs the new wave of robotics that learns actions from data.
More from this episode
Jason explains that traditionally you have robots as an automation tool — you see them in factories, pre-programmed with an exact sequence of motions, like packaging things into a box to deliver. That's very difficult for something like folding laundry, "because your clothes is like deformable, the shapes are always different, they're not in perfect rigid shapes and states." So it's really hard to pre-program a sequence of motion to fold cloth. What you need instead is to develop AI models — much like how a language model or ChatGPT can do anything based on your language command — that can control robots to do very fluid and dynamic motions, just like how human arms are fluid and dynamic. That's why the task was very difficult for the preceding paradigm of robotics and is now more amenable to the new wave of robotics, which is learning robot actions through data.