Ad

Friday, September 1, 2017

Roomba and Machine Learning Artificial Intelligence for the cleaning robot



The latest Roomba 980 can scan room size, identify obstacles and optimize routes. This robot is equipped with state of the art sensors and cameras to scan the room, scan for obstruction and record odometry (used by wheeled robot to estimate distance traveled from a starting position). Its decision tree may work something like scan a small room 3x, a medium size room 2x, a large room for only once. source

It is equipped with infrared receiver for sensors such as cliff sensors and object sensors. It calculates the room size based on distance traveled. The wall sensor allows iRobot to travel closely along the wall.  The iRobot 980 camera can look forward and up at a 45 degree angle.



In machine learning, robot - the agent will interact with the environment, record each state and reward for moving into the next date, update a Q table with utility and eventually choose the best way to complete the task. Reinforcement learning such as that allows the robot to optimize its routes and maximize reward for completing a task based on a set of policies.

In reality, Roombas are forgetful (resets after each run) but it's getting advanced AI functionalities fast. With its existing cameras and image processing software, Roomba iRobot can map out your room with surprising precision. The camera and software in the  Roomba iRobot 980 device can navigate much better than its predecessors which move around semi randomly (at one point Roomba Red travels in spirals, the SPOT cleaning feature still looks a bit like that). 980 has vision! It does not recognize objects yet.

Roomba uses  simultaneous location and mapping or SLAM, an algorithm that takes significant time to optimize and is a lot to pack into a small device according to researchers at MIT. MIT professor John Leonard says Google self driving cars already use navigation systems based on SLAM technology (the self driving car also made significant improvements and use a whole lot more data than SLAM for iRobot which a simple localization task only source).

This little robot is mapping out your room. With the newest Roomba connecting to WIFI and working with Alexa and Google Homes, researchers are concerned about data collection and privacy. The user has to keep Roomba offline or explicitly opt out of data sharing for the advanced wireless models. Albert Gidari, director of privacy at the Stanford Center for Internet and Society, told NYTimes that sharing such data will draw legal ramifications.
source

Did you know that iRobot was created by MIT alumni?

Roomba reached 655 millions in sales 2016. (source)
https://uniqtech.tumblr.com/post/164869166245/irobot-roomba-980-algorithm-machine-learning-slam

No comments:

Post a Comment

Rules of Sudoku for Algorithm Exercises

Need to code a Sudoku solver? Here are three rules of Sudoku: A 9x9 grids, Each row ... Each column ... Each of the 9 3x3 grids (examp...