Keywords: Partial Order Planning, Object Scouting, Locally Observable MDP
Abstract: State uncertainty is one of the primary obstacles to
effective long-horizon task planning in robotics. We tackle state
uncertainty by decomposing it into spatial uncertainty—resolved
by SLAM—and uncertainty about the objects in the environment,
formalized as the object scouting problem and modelled using
the Locally Observable Markov Decision Process (LOMDP). We
introduce a new planning framework called Scouting Partial
Order Planner (SPOP), specifically designed for object scouting
with LOMDPs, which exploits the characteristics of partial order
and regression planning to plan around gaps in knowledge the
robot may have about the existence, location, and state of relevant
objects in its environment. Our preliminary results demonstrate
how the aspects of partial order planning make it uniquely suited
to solving tasks in the object scouting context: SPOP significantly
outperforms alternative planners in this setting.
Submission Number: 48
Loading