Probabilistic backward and forward reasoning in stochastic relational worlds
Tobias Lang and Marc Toussaint
In: ICML 2010(2010).
Inference in graphical models has emerged
as a promising technique for planning. A
recent approach to decision-theoretic plan-
ning in relational domains uses forward infer-
ence in dynamic Bayesian networks compiled
from learned probabilistic relational rules.
Inspired by work in non-relational domains
with small state spaces, we derive a back-
propagation method for such nets in rela-
tional domains starting from a goal state mix-
ture distribution. We combine this with for-
ward reasoning in a bidirectional two-lter
approach. We perform experiments in a
complex 3D simulated desktop environment
with an articulated manipulator and realis-
tic physics. Empirical results show that bidi-
rectional probabilistic reasoning can lead to
more ecient and accurate planning in com-
parison to pure forward reasoning.