In this work, the problem of inferring the time-evolving, 'hidden' state of a 
dynamical system via adaptive exploitation of heterogeneous sensing modalities
is considered. Building on our prior unified framework of tracking and
control,
our focus is on enhancing the quality of system state estimates by exploiting
both past and future observations and controls. Following an innovations
approach, Kalman-like smoothers are designed to obtain more refined system
state
estimates of the Markov chain system state. Numerical results are also
presented
for the application of physical activity detection in Wireless Body Area
networks. The power of the proposed framework is its ability to accommodate a
plethora of controlled sensing applications.