00:00Lianjian Su, Chief Analyst at Technology Market Research Company, Omdia,
00:04says new AI world models are vital to upskilling robots.
00:10The more sort of advanced way of deploying world models is to actually compress the model
00:15and then deploy onto the robot itself.
00:18So basically very similar to what most of the human robotics vendors are doing.
00:23So having a model situated on the robot physical body
00:29and then that will dramatically reduce the latency.
00:32It will allow the robot to react faster or more accurate to the surrounding.
00:37But that means also that the robot needs to have much higher processing capability
00:41than what it has today.
00:43So we're hearing these terms spatial intelligence and real-to-sim workflows.
00:48Do you think is this the new AI frontier?
00:51It is definitely the new frontier for physical AI.
00:54So anything that we want an AI to interact with the physical environment or physical object,
01:02you will need robots to migrate very seamlessly from simulation into a real-life environment.
01:10And then you also want the robot to truly understand the relationship of the items that it sees
01:18or the object that it's perceived with its grandeur surrounding or its ambient environment.
01:26That relationship, that understanding is hard to acquire in the older days of AI.
01:33And with world model, the idea is that you will speed up that understanding.
01:36You will have a much better perception and navigation capability.
01:43Now, World Labs is not the only company, of course, working on this.
01:47How fierce is the competition and who's putting in the money?
01:50It is quite fierce.
01:52So World Labs obviously is headed by Fei-Fei Li,
01:56which is obviously a very prominent AI scientist in the world.
01:59You have direct competitors in the Western world.
02:02So AMI Labs, headed by Ian Lee Kuhn, ex-meta AI scientist,
02:06which is also producing a fairly good version of world models.
02:11But if you cast your eye across the world, there are different countries working on world models.
02:16Obviously, the other prominent example is China.
02:19You have SenseTime, for example, having their ASI Robotics subsidiary working on their version of world models
02:26that is deployed directly onto the robot itself.
02:29So that market, I think, or rather this world model market is going to gain significant momentum in the next
02:36few months,
02:37especially when the race behind physical AI started to really heat up.
