This talk presents new results on robots collaboratively controlled by humans
via networks such as the Internet. The Tele-Actor is a system that combines
a human agent with distributed audience control for applications ranging from
distance learning to journalism to entertainment and gaming. I'll describe
our experience with a series of field tests, architectures, technologies,
and applications, including an unsupervised scoring algorithm that automatically
computes and rewards "leadership" behavior among players.
In the second half of the talk, I'll discussed networked robotic cameras.
Newly available robotic cameras offer pan, tilt, and extreme zoom capabilities
with built-in network servers at low cost. I'll describe a recent project
where we installed such a camera at Sproul Plaza at UC Berkeley to raise questions
about privacy in public spaces. Such camera motivate the Single Frame Selection
(SFS) problem, where n users share control of a single robotic camera.
I'll present several algorithms, O(n^2 m) for a set of m zoom
levels, and O((n + 1/\epsilon^3) log^2 n) for an infinite set of zoom
levels. The algorithms can be distributed to run in O(n m) time at
each client and in O(n \log n) time at the server.
Dez Song, Judith Donath, Sariel Har-Peled, Vladlen Koltun, and Frank van
der Stappen have contributed to this work.
Ken Goldberg
is Professor of IEOR and EECS at UC Berkeley. His research addresses robot
manipulation, geometric algorithms for automation, and networked robots. More
information and online projects are linked from www.ieor.berkeley.edu/~goldberg.