Software-Architectures for Agents and Mobile Robots 


Agents and mobile robots are implemented to act "autonomously on behalf
of their user/owner". They have to interact with virtual or real-world 
environments. This leads to a first "horizontal" modularization according 
to perception, control, and actuation. Reactive behavior is implemented 
by simple translations from sensors to actuators, deliberative behavior 
includes complex goal selection and planning. Hybrid architectures combine
both approaches using layered architectures, which leads to a second 
vertical modularization. The synchronization and interaction between the 
modules poses serious problems when the agents/robots have to work on 
complex  tasks in dynamic environments. Persistent states are used to 
maintain past oriented and future oriented information: The world model 
combines new perceptions with previous ones, and the commitment maintains 
plans for the achievements of long term goals. Special efforts are needed 
to keep balance between stabile behavior and adaptation to new situations.
The implementation of "bounded rationality" needs new architectures behind
the scope of the classical ones.