AI Paper Atlas

Research map / Reinforcement learning / MDPs and planning

Policy-Conditioned Uncertainty Sets for Robust Markov Decision Processes

Andrea Tirinzoni, Marek Petrik, Xiangli Chen, Brian Ziebart

NeurIPS 2018 · Reinforcement learning › MDPs and planning · approach: Robust MDP policy gradient · 9 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use marginally-constrained uncertainty sets instead of s,a-rectangular setsRobust MDP policy · Invasive species, M=50, p=0.3negative expected return: 117.15 ± 6.76 -> 98.36 ± 0.78Improved
use marginally-constrained uncertainty sets instead of s,a-rectangular setsRobust MDP policy · Invasive species, M=100, p=0.5negative expected return: 136.24 ± 8.41 -> 115.04 ± 6.18Improved
use marginally-constrained approach instead of MLE dynamicsPolicy from estimated dynamics · Invasive species, M=50, p=0.3negative expected return: 140.36 ± 1.28 -> 98.36 ± 0.78Improved
use marginally-constrained approach instead of fitted Q-iterationBatch policy learning · Gridworld, large gridsexpected return under true dynamicsImproved
use marginally-constrained approach instead of s,a-rectangular uncertaintyRobust policy · Gridworld, increasing grid sizeexpected return under true dynamicsImproved
rectangular vs marginal constraints on small grids (N <= 7)All methods · Gridworld, small gridsexpected returnNo clear effect
increase number of discrete belief states (ablation)MC dynamic program · 5x5 gridworldapproximation errorImproved
increase trajectories from 50 to 100 (ablation)MC · Invasive species, p=0.3negative expected return: 98.36 ± 0.78 -> 96.20 ± 0.95Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy