Crowdsourcing robot training to make robots learn faster and smarter

Some studies by Conor, MIT and the University of Washington have shown that robots are expected to learn what humans want them to do, but only if they have to provide enough data for training.

The University of Washington Research program teaches robots to build things with Lego blocks according to human examples. In order to improve accuracy, the researchers asked the robot to analyze various examples of specific objects (people or turtles) that were packed out, and found that robots tended to choose between simple and original similarities.

MIT's research focuses on finding solutions to the facts that are available to a number of robots (or any node in a distributed system) in order to realize the collective intelligence of the robot. One way to do this is to have multiple robots investigate the same building, and then classify the space according to what's inside. It may be different for each robot in the same space, but by constantly comparing it to each other, these robots will eventually be able to deduce the exact model together.

Cornell University Research is an interactive online demo called Tell Me Dave. The demo teaches the robot how to perform certain actions, even those that do not indicate details. Researchers will use special commands to train robots (such as taking a pot to the stove), and researchers will only tell the robot what the kettle looks like and what it is for. For example, to tell a robot to "boil water", the robot will know how to boil water in a microwave oven or stove, and know that the stove switch must be hit. The robot's training is done by letting the volunteer play the first person game to perform the task.

The process of teaching robots is very similar to teaching children, because this is a system of learning, but learning needs to teach, while teaching robots requires a lot of data, so machine learning and large data is the key to the practical skills of the robot. Crowdsourcing is clearly an effective way to get data from robots, and to help the latter learn things faster by the collective power of humans and robots. Perhaps the robots that can really understand human natural language and perform tasks are not far away.