To teach robots to understand spoken language, Cornell University is trying to give up command.

Cornell University is conducting a study at the school's Robotics lab, which aims to teach robots to understand popular voice commands without the need for explicit and complex instruction strings.

The experiment was initiated by a professor and two doctoral students. Their basic idea is to teach robots to listen to and understand basic spoken instructions from different speakers, and then to "adapt to the environment" and fill in the missing information content.

I know you look like you're going to have a hard case. When the user says, "Put the teapot under the faucet and fill it with water, then turn off the faucet and boil the kettle," the robot will consider the position of the teapot and the faucet, how to move the object to the target location, and the understanding of where to boil the heat.

But the research is not perfect, and with enough training, the robot has been able to feed back the correct action in 64% of the time. When an order or environment changes, it can even learn and change behavior.