The work drsrrihrd in this thesis ia about, how to program robots to work re
liably in the presence of uncertainty. Some architectural principle!: are proposed
which address the problem of decomposing robotic assembly tasks into modular
units such that a robot program can be implemented efficiently, tested easily, and
can be maintained or modified without undue complexity. This architecture also
provides a framework to integrate sensors into a robotic, assembly system.
These modular units arc called behavioural modules. They perforin their tasks
reliably. The problem of uncertainty is dealt with by encapsulating sensing and
variation reducing strategies inside these modules. Experiments are performed with
a working robotic assembly system using vision based behavioural modules. Analy
sis of this system validates the principles presented in this thesis.