| |
Abstract:
A probabilistic algorithm is presented for learning the
dynamics of complex motions. Complexity is taken into account by
allowing multiple classes of motion, and an Auto-Regressive Process
(ARP) associated with each class. Training sets need incorporate
only indirect observations of motion, and this allows for sensory
noise. A learning algorithm is presented for this problem based on
propagation of random samples. Experiments have been performed with
visually observed juggling, and plausible dynamical models are
found to emerge from the learning process.
|
