The history I-space is simply the set of all possible history
I-states. Although the history I-states appear to be quite
complicated, it is helpful to think of them abstractly as points in a
new space. To define the set of all possible history I-states, the
sets of all initial conditions, actions, and observations must be
precisely defined.
The set of all observation histories is denoted as
and
is obtained by a Cartesian product of copies of the observation
space:
|
(11.16) |
Similarly, the set of all action histories is
, the
Cartesian product of copies of the action space .
It is slightly more complicated to define the set of all possible
initial conditions because three different types of initial conditions
are possible. Let
denote the initial condition space.
Depending on which of the three types of initial conditions are used,
one of the following three definitions of
is used:
- Known State: If the initial state, , is given, then
. Typically,
; however, it might be known
in some instances that certain initial states are impossible.
Therefore, it is generally written that
.
- Nondeterministic: If is given, then
(the power set of ). Again, a typical situation is
; however, it might be known that certain subsets of are
impossible as initial conditions.
- Probabilistic: Finally, if is given, then
, in which
is the set of all probability
distributions over .
The history I-space at stage is expressed as
|
(11.17) |
Each
yields an initial condition, an action
history, and an observation history. It will be convenient to
consider I-spaces that do not depend on . This will be defined by
taking a union (be careful not to mistakenly think of this
construction as a Cartesian product). If there are stages, then
the history I-space is
|
(11.18) |
Most often, the number of stages is not fixed. In this case,
is defined to be the union of
over all
:
|
(11.19) |
This construction is related to the state space obtained for
time-varying motion planning in Section 7.1. The
history I-space is stage-dependent because information accumulates
over time. In the discrete model, the reference to time is only
implicit through the use of stages. Therefore, stage-dependent
I-spaces are defined. Taking the union of all of these is similar to
the state space that was formed in Section 7.1 by making
time be one axis of the state space. For the history I-space,
, the stage index can be imagined as an ``axis.''
One immediate concern regarding the history I-space
is
that its I-states may be arbitrarily long because the history grows
linearly with the number of stages. For now, it is helpful to imagine
abstractly as another kind of state space, without paying
close attention to how complicated each
may be to
represent. In many contexts, there are ways to simplify the I-space.
This is the topic of Section 11.2.
Steven M LaValle
2020-08-14