00:03We've all fantasized about a robot doing our household chores, haven't we?
00:08But one Chinese company says that could soon be a reality.
00:11The CEO of Unix AI, Fred Yang, predicts the market for such products could rival that of smartphones.
00:18He's been speaking to CGChain's Li Chuan Hua about the technology behind it, known as Unitouch.
00:24So basically, in the previous age of robotic manipulation, people only used cameras, 2D cameras or 3D cameras,
00:35for capturing some videos for guidance to guide where the robotic arm should be.
00:42You cannot tell how much force you need to press or pick something using only the vision signal.
00:48So tactile signal is basically, we mimic how the human are oppressing the object
00:56and the object is giving a counter feedback to the human finger itself.
01:01Actually, the function of the Unitouch is that we use vision as an intermediate modality
01:07that actually connects tactile to vision, language, audio and other modalities
01:14that may not be naturally paired with tactile signals.
01:17To be concise, after we have this model, we are building the connection between tactile signal and vision signal,
01:23and thus we can generate tactile signal even without touching the object.
01:28Can I say that your design philosophy is utility or practicality over humanoid?
01:34I would say yes.
01:36So the reason we have this foundation model is that you will find that actually tactile sensor
01:42is a very immature sensor compared to the vision, compared to the camera or other sensors.
01:50It's pretty expensive.
01:52So when I was a PhD at Yale just two years before, the cost of one tactile sensor is roughly
01:59$1,000.
02:01And there's a gel on the top of the sensor.
02:04Normally, that can only last for three to five years.
02:07And that makes the tactile signal and tactile sensor really hard to commercialize.
02:12So my design philosophy for this model is that can we use the tactile signal without this very expensive
02:23but very inelastic tactile sensor?
02:30So this is the philosophy behind the model.
02:32For today, a humanoid robot is a very wide conceptual definition.
02:37It's not only the robot that has two legs and two arms that we call the humanoid robot,
02:43but also the robot that actually works like a human, we call it embodied intelligence, a humanoid robot.
02:50And how big is the market for embodied artificial intelligence you mentioned just now?
02:56So in theory, it's a very large one.
03:00It's a countless market.
03:02But I think for now, we are only focusing on a very, very small subset of the market.
03:07But even though that is a very huge market.
03:10So take one of the jobs we are doing, for example.
03:14For example, we have a robot that can act as a safeguard in schools, buildings, hospitals, and home and houses.
03:26Merely in China, we have, I think, 8 to 10 millions of people doing this kind of job.
03:32This is only one of the very small pieces of the task that a human robot can do.
03:39So in general, I would say the embodied intelligence market is a countless market.
03:43Are you betting on a future that robots will be like smartphones, that everyone will have one in their households?
03:51So actually, I even have a bet that it's larger than a smartphone.
03:55So for smartphones, it's that everyone has a robot.
03:59Only is it to see, is it to customer side as a consumer business.
04:05But for a robot, if we are going to the ultimate phase,
04:08it's not only an item that every people should have in their house,
04:13it also enters into industry.
04:15It enters into the service field.
04:17It also enters into every field that actually a human is working right now.
Comments