Artificial Intelligence and Bayes Rule
Bayes Rule is a prominent principle used in artificial intelligence to calculate the probability of a robot’s next steps given the steps the robot has already executed. PR2, the newly formed coffee making robot, can make coffee with any coffee machine if the user gives it a list of instructions to follow. The robot’s uniqueness lies in its ability to operate autonomously using the same intuitive extrapolation skills that a human would use. An average person has a lot of appliances in their house that they have been used to operating such as a stove, toilet, or door knob, so that even if they encounter an unfamiliar appliance, they can generalize how to use it based on their familiarity with similar appliances. Similarly, the PR2 transfers trajectories from things it has experienced to unfamiliar territory. For example, the robot identifies all the parts on the coffee machine and searches a database that contains the known methods for operating similar parts. Consequently, the robot can apply the mechanism for operating a toilet or a lever into operating the hot water nozzle on an espresso machine.
PR2 Reading Instructions to make Coffee
Bayes rule helps the robot in deciding how it should update its knowledge based on a new piece of evidence. Since the PR2 is capable of operating on any new coffee machine, it changes its operating function (i.e. the way it touches, pushes, or grabs objects ) through what it sees presented right in front of it. Suppose that a new coffee machine has a button to release the hot water while the older machine had a lever. In order to calculate the probability that the robot will be able to push the button successfully given that it has only operated on levers before will be determined by Bayes Rule. Given that probability, the researchers develop an algorithm that will allow the robot to search through a database of different appliances if the probability of pushing the button is very low. This allows the robot to get familiar with buttons in other appliances and increases its chances of actually pushing the button. This newly acquired skill will help the robot operate on similar machines that have buttons in the future. Essentially, the robot is analyzing the evidence that it is presented with, comparing it to its previous tasks, calculating the probability of a successful execution and then searching through a database to find the most suitable trajectory for its task. Bayes rule allows the robot to check unfamiliar trajectories with known trajectories , so that it can optimize its performance for the user.
Link: http://spectrum.ieee.org/automaton/robotics/artificial-intelligence/pr2-robot-latte