<s>
Constructing	B-Algorithm
skill	I-Algorithm
trees	I-Algorithm
(	O
CST	O
)	O
is	O
a	O
hierarchical	O
reinforcement	O
learning	O
algorithm	O
which	O
can	O
build	O
skill	O
trees	O
from	O
a	O
set	O
of	O
sample	O
solution	O
trajectories	O
obtained	O
from	O
demonstration	O
.	O
</s>
<s>
CST	O
uses	O
an	O
incremental	O
MAP	O
(	O
maximum	B-General_Concept
a	I-General_Concept
posteriori	I-General_Concept
)	O
change	O
point	O
detection	O
algorithm	O
to	O
segment	O
each	O
demonstration	O
trajectory	O
into	O
skills	O
and	O
integrate	O
the	O
results	O
into	O
a	O
skill	O
tree	O
.	O
</s>
<s>
A	O
particle	B-Algorithm
filter	I-Algorithm
is	O
used	O
to	O
control	O
the	O
computational	O
complexity	O
of	O
CST	O
.	O
</s>
<s>
Then	O
,	O
CST	O
compute	O
the	O
probability	O
of	O
the	O
changepoint	O
at	O
time	O
with	O
model	O
,	O
and	O
using	O
a	O
Viterbi	B-Algorithm
algorithm	I-Algorithm
.	O
</s>
<s>
The	O
following	O
pseudocode	B-Language
describes	O
the	O
change	O
point	O
detection	O
algorithm	O
:	O
</s>
<s>
CST	O
is	O
much	O
faster	O
learning	O
algorithm	O
than	O
skill	B-Algorithm
chaining	I-Algorithm
.	O
</s>
<s>
CST	O
has	O
been	O
used	O
to	O
acquire	O
skills	O
from	O
human	O
demonstration	O
in	O
the	O
PinBall	B-Application
domain	O
.	O
</s>
