This is a really good question because this is what actually separates artificial intelligence from robotics!
A typical robot that is used in factories only performs tasks that you programmed it to do. You can program it to react to its environment using sensors or vision systems, but all the actions are programmed into it.
On the other hand, if you teach your robot via Machine Learning (an approach to achieve artificial intelligence), you can be surprised by what the robot is doing!
There is a category of machine learning called “reinforcement learning” where you don’t program the trajectories, you only give the robot an objective to achieve. You also define a rewarding system – just like training your dog to do something! If you want your dog to give one paw when you call its name, you reward him/her with a treat when the dog actually does give its paw and don’t give any reward when the dog doesn’t. After a lot of training, the dog will understand that by giving its paw, he/she will receive a treat. You can teach a robot using a similar method! And in that case, the robot will act “by itself” to reach its objective and to get its reward.
There is a lot of different machine learning applications that you could use for robotics, it is a really interesting field!
Comments
Curious commented on :
Thanks Sabine,I voted for you