0.25
0.5
0.75
1.25
1.5
1.75
2
Efficient Policy Construction for MDPs Represented in Probabilistic PDDL
Published on Jul 21, 20113199 Views
We present a novel dynamic programming approach to computing optimal policies for Markov Decision Processes compactly represented in grounded Probabilistic PDDL. Unlike other approaches, which use
Related categories
Chapter list
Policy Construction for MDPs Represented in Probabilistic PDDL00:00
Outline00:01
Motivations00:26
Compact Action and Value Function Representation01:33
PPDDL at a glance02:06
Policy construction - RBAB Algorithm (1)03:24
Frameless Action-Value Functions (F-Values)03:28
Why F-Values ?04:40
From PPDDL Effects to F-Values05:30
Policy construction - RBAB Algorithm (2)05:56
Example Action Backup06:03
F-Value for an update effect06:55
F-Value for a simple effect07:10
F-Value for a conditional effect07:53
F-Value for a probabilistic effect08:45
From F-values to action values09:29
Value Iteration with F-Values10:30
Policy construction - RBAB Algorithm (3)11:02
Evaluation on IPC Domains (1)11:06
Evaluation on IPC Domains (2)12:30
Evaluation on IPC Domains (3)13:16
Evaluation on IPC Domains (4)13:37
Policy Revision With F-values13:53
A Policy Revision Problem13:57
Possible Revisions14:29
Conclusion15:47
Thank You17:25