Manuela Veloso, Herbert A. Simon University Professor, Computer Science Department, Carnegie Mellon University.
Robots use sensors and reasoning to have autonomy. They use radars, lasers and cameras to see, and wheels to get around. There are Roombas, but in general robots that get around by themselves are rare.
Robots can learn a map of a building, or calculate it based off a PDF – and uses it’s own motion model. Check out Depth-Camera Based Indoor Robot Localization , Joydeep Biswas, PhD 2014 for more details.
Kinect can be used to play games, but can be used by robotics. It can perceive color and distance. We sample the point cloud, compute the normal of three points to define a plane. Do this many time to do the plane filters. Why do we care? Because walls are planar, floors are planar. We got to see a cool video of the robot processing data (from it’s perspective).
CoBot Learning Data, Richard Wang, PhD 2016 about how to use a robot to gather accurate data about wifi, temperature, etc from Robot Localization.
As smart as these robots are, they still cnnot really go up stairs, open all doors, etc. Some do not have arms – so they cannot put elevator buttons. Now they have to ask for help! Symbiotic Autonomy! They can ask for help from humans, the Internet, or other robots.
CaBot at CMU needs to ask a human to push the elevator call button, tell it when the elevator has arrived and which one it is (and then push the button for it’s destination floor).
But, what if nobody helps? CaBot will send email to remote humans – “I have been blocked by an obstacle for more than 3 minutes. Can someone please come to office 7409 on floor GHC7 to rescue me?”
Only happens a few times a year, most people will move out of the way for the robot or help it out.
Human-Centered Planning, Stephanie Rosenthal, PhD 2012 – could human planning be part of the robot’s planning? That is, choosing routes with more helpful humans, though not necessarily the shortest Dr. Veloso no longer gives people directions to her office – she sends CoBot to meet and guide them.
The robot has to look up references to objects it may not have been trained on – like where to find coffee or chocolate, or what a binder is? Can look up on the internet to calculate the most likely place to find a particular object. Watched a fun video of the robot seeking coffee to take to the lab, which of course, as it has no arms, required human assistance. It will learn from that information and now in the future it will know where to find coffee.
At CMU, they had 4 robots… but one professor moved to UMass, so now UMass has one CoBot and CMU has 3. 🙂
Videos of robot soccer! These robots are not remote controlled, nor playing a programmed game – they are dependent on teamwork to complete the task.
The CoBot robots talk to each other, to help them plan and replan based on closed doors and obstacles. They are sharing information!
Brian Coltin, PhD 2014 did research for multi-robot transportation task planning with transfers. His research also tracked taxi trips – if taxis were autonomous and could transfer their passengers to one cab at common crossing points so only one taxi went to the airport, lots of time and fuel could be saved.
CMU has another robot, Baxter, who has arms, but cannot move around. CoBot can move, but has no arms – but does have a basket. So, the two robots can work together 🙂
I’ve already seen robots at work in the Mountain View, CA, El Camino hospital, delivering medications and moving around paperwork, linens and trash. 🙂