The ability to learn actions from human demonstrations is one of the major challenges for the development of intelligent systems. However, researchers from the University of Maryland and NICTA, Australia have successfully taught robots how to cook after watching YouTube tutorials.
The humanoid robots are able to learn specific grasping actions required to manipulate kitchen tools and then determine the most efficient combination of motions allowing them to complete specific tasks.
The system relies on a type of artificial intelligence called ‘deep learning artificial neural networks’ to train the robots using a variety of information from videos, audio and images gathered from 88 YouTube videos. Although this type of information is easy for us to process and understand, it is far from simple for an artificial system.
These robots made use of recent advances in computer vision, natural language processing and increased computational power to allow them to make sense of the data.
This research is a crucial step towards robots understanding what humans are doing. Whilst previous work has tried to develop artificial intelligence systems that imitate the movements of humans, here the system tries to copy the goals allowing the robot to decide for itself what the best combination of actions would be rather than simply reproducing a series of behaviours.
Ultimately, the aim of robotics research such as this is to develop a fully intelligent robot for manipulation tasks which can automatically enrich its own knowledge by extracting information from the internet. This would have implications far beyond kitchen as humanoid robots could adopt a wide range of roles in society from cleaning to construction saving both time and money. Whilst this may seem a daunting prospect for the job market, the trend towards increasing automation in all areas of society seems like an inevitability we will have to accept.