AI Paper Atlas

Research map / Reinforcement learning / Policy gradient and actor-critic

Predictor-Corrector Policy Optimization

Ching-An Cheng, Xinyan Yan, Nathan Ratliff, Byron Boots

ICML 2019 · Reinforcement learning › Policy gradient and actor-critic · approach: Alternative policy update rules · 12 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 PicCoLO prediction and error-correction steps instead of the model-free updatefirst-order policy optimization · Predictable online learning (theory)average regret upper bound: \frac{1}{\sqrt{N}} \paren{\abs{\Pi} + G_g^2 + \sigma_g^2} -> \frac{1}{\sqrt{N}} \paren{\abs{\Pi} + \sigma_g^2 + \sigma_{Improved
Use an accurate predictive modelPicCoLO with Adam · CartPolesum of rewardsImproved
Use an adversarial predictive model (ablation)PicCoLO with Adam · CartPolesum of rewardsNo clear effect
Use an adversarial predictive model (ablation)Dyna with Adam · CartPolesum of rewardsHurt
Compute the predicted gradient with the fixed-point formulation rather than Φ(π̂_n) (ablation)PicCoLO · CartPolesum of rewardsImproved
Compute the predicted gradient with the fixed-point formulation rather than Φ(π̂_n) (ablation)PicCoLO · Robot RL taskssum of rewardsImproved
PicCoLO a base optimizer using predictive gradientsAdam, Natgrad, and trpo · Hopper, Snake, and Walker3Dsum of rewardsImproved
Use predictive gradients from a biased dynamics model with fixed-point prediction (ablation)PicCoLO · Robot RL taskssum of rewardsMixed

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