MIT researchers have developed a system to teach robots context. This allowed a test robot to correctly process ninety percent of complex commands.
Difficult to communicate in human language
It is difficult to communicate with a robot in human language. Much of human communication has contextual references. For a robot, this is not so easy. In order to solve this problem, a research group developed the ComText system for MIT.
Robot with context
If you instruct a robot from nowhere to 'pick up my apple', he will not understand what you mean. Tell him first 'the green apple I put down here is my apple' and ask him again afterwards, then he will - thanks to ComText - know what to do. By giving the robot more and more information, the system is updated and he can understand more and more commands with contextual references. ComText was tested on the two-armed Baxter robot, which managed to correctly execute ninety percent of such instructions.
The MIT works with the so-called Episodic memory. Episodic memory is the memory for personal events and also relevant for machine learning. Together with the general facts and figures (semantic memory) available in the cloud, this enables robots to perform even better.