Autonomously seeking out power for battery recharging is a pretty crucial capability for advanced mobile robots. While Roomba-like docking stations are a quick fix, "plugging in" to existing infrastructures is preferable. Not long ago, the robotics world was abuzz with the Willow Garage Milestone 2, where (among other things) a PR-2 robot plugged itself into 9 different wall outlets. My curiosity on this subject was further piqued when I saw Intel's Marvin robot use electric fields emanating from an outlet's internal wiring to finely localize an outlet/plug and adeptly plug itself in, all sans camera. I'd like to share some photos and videos of recent efforts (by both the Willow and Intel folks), as well as examine the history of robots plugging themselves into wall outlets.
Going in reverse chronological order, lets take a look at the latest Intel Research Lab efforts first.
Intel's mobile manipulating robot, named Marvin, appears to be a Segway RMP 100 with Barrett WAM Arm and runs the Willow Garage ROS NavStack. The thing that makes this work really cool is the power plug, which is specially 3D-printed for the robot's end-effector and contains integrated electric field (E-Field) sensing electronics. The plug has antennas (electrodes) that sense the 60Hz electric fields emanating from a powered outlet in order to facilitate highly-accurate plug-outlet localization and subsequent plugging-in. Take a look at the photos and videos to get a better idea of what I mean.
One of the reasons this is so clever is that the plug-outlet localization utilizes a signal that is explicitly created by the target, much in the spirit of Shwetak Patel's "infrastructure mediated sensing." This is in contrast to the visual methods employed by most other researchers. Plus, the circuitry is rather straight forward (see below left) and produces amazingly accurate localization (see below right). Thankfully, the originator of this work (Brian Mayton) has shared an ICRA 2010 pre-print that details the solution.
An alternative approach, championed by Willow Garage and their PR-2 robots back in June 2009, is to employ visual techniques to facilitate plug-outlet localization. Plugging in was part of their wildly successful Milestone 2, where the PR-2 autonomously plugged itself into 9 different office outlets. While early versions relied on a visual fiducials on the plug (checkboard pattern) and worked only on orange outlets (though quickly extended to many others), I'm sure subsequent efforts are making inroads into more robust / generalized behaviors. Personally, I'm looking forward to the detailed systems papers to be presented at ICRA 2010, as promised by Gary Bradski. Anyway, Willow has videos of Milestone 2 and an explanation of how it was accomplished -- photos and videos embedded below.
Now for some historical context... Amazingly enough, the first robots capable of plugging themselves into wall outlets were developed in 1960 -- the Johns Hopkins "Beast" Robot:
Controlled by dozens of transistors, the Johns Hopkins University Applied Physics Lab's "Beast" wandered white hallways, centering by sonar, until its batteries ran low. Then it would seek black wall outlets with special photocell optics, and plug itself in by feel with its special recharging arm. After feeding, it would resume patrolling. Much more complex than Elsie, the Beast's deliberate coordinated actions can be compared to the bacteria hunting behaviors of large nucleated cells like paramecia or amoebae.
A quality treatise on the subject of robots plugging themselves in to wall outlets is Eduardo Torres-Jara's 2002 MIT Masters thesis, entitled "A Self-Feeding Robot." Curiously, his robot bears a striking resemblance to the Beast robots of old, except that his used a two DoF linear actuator "arm" to plug itself in and located / localized outlets using boosted visual classifiers.
While Roomba-like docking stations are an option for low-cost commercial robots, I believe that autonomously plugging into unmodified power outlets for recharging is a baseline capability for advanced home/office robots. Non-roboticists certainly marginalize the difficulty of this fundamental capability; however, having hand coded a number of robot behaviors myself, I can certainly appreciate the capabilities recently demonstrated by researchers -- kudos guys!