Wireless sensor networks are often di±cult to program and
unable to adapt to a changing environment. Mobile agent
middleware promises to address both concerns by providing
higher-level programming abstractions and the ability to in-
ject new agents into a preexisting network. The unique char-
acteristics of wireless sensor networks like resource scarcity
and emphasis on spatial locality require new algorithms for
controlling agent behavior. This paper presents a procedure
for one speci¯c behavior: network exploration. Network
exploration is needed by many tasks ranging from simple
data collection to network health monitoring. Our proposed
procedure uses a genetic algorithm to determine the num-
ber of agents and their itineraries, followed by techniques
for in-network adaptation to unpredictable situations like
node failure. This paper presents a genetic algorithm and
its adaptation strategies. The procedure is evaluated using
a wireless sensor network consisting of 25 Mica2 motes run-
ning Agilla, a mobile agent middleware for wireless sensor
networks.
