We present an active vision algorithm for computing the orientation and pos
ition of a locally planar object, onto which is cast a shadow of the edge o
f a half-plane at an unknown location. This algorithm utilises active posit
ion control of a point light source, and employs a Kalman filter to perform
temporal integration of measurements. The light source position is adjuste
d after each measurement so as to reduce the trace of the expected state es
timate error covariance matrix for the next measurement. We demonstrate the
active shape-from-shadows algorithm using a real robotic system.