Most of the cover of humanoid robotics has understood to focus on material design. Given the frequency with which their developers throw around the expression “humanoids for general use”, more attention should be paid to the first bit. After decades of single -use systems, leap to more generalized systems will be important. We are not there yet.
The push to produce a robotic intelligence which can take full advantage of the extent of the movements opened by the biped humanoid design was a key subject for researchers. The use of the generative AI in robotics was also a subject heated to white. New research Out of Mit underlines how the latter could deeply affect the first.
One of the biggest challenges on the road to general use systems is training. We have a good understanding of the best practices to train humans how to make different jobs. The approaches to robotics, although promising, are fragmented. There are many promising methods, including learning to strengthen and imitation, but future solutions will likely involve combinations of these methods, increased by generative AI models.
One of the main use cases suggested by the MIT team is the possibility of gathering the relevant information from these small sets of specific data. The method was nicknamed the composition of policies (POCO). The tasks include useful robot actions such as hammering in a nail and turning things with a spatula.
“(The researchers) form a distinct diffusion model to learn a strategy or a policy, to accomplish a task using a specific data set,” notes the school. “Then, they combine the policies learned by the diffusion models in a general policy which allows a robot to perform several tasks in various contexts.”
By MIT, the incorporation of diffusion models has improved the 20%task performance. This includes the possibility of performing tasks that require several tools, as well as learning / adaptation to unknown tasks. The system is able to combine relevant information from different data sets in a chain of actions necessary to perform a task.
“One of the advantages of this approach is that we can combine politicians to make the most of the two worlds,” said the main author of the newspaper, Lirui Wang. “For example, a policy trained on real world data could be able to achieve more dexterity, while a policy formed in simulation could be able to achieve more generalization.”
The objective of this specific work is the creation of intelligence systems that allow robots to exchange different tools to perform different tasks. The proliferation of versatile systems would bring the dream industry closer to general use.