Research map / Reinforcement learning / Policy gradient and actor-critic
Alternating Optimisation and Quadrature for Robust Control
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| model both policy and environment variables, then select policies with BO and environment settings with BQ | ALOQ · Robotic arm, Collision Avoidance | expected cost: 8.9 vs 28.3 | Helped |
| model both policy and environment variables, then select policies with BO and environment settings with BQ | ALOQ · Robotic arm, Joint Breakage | expected cost: 7.7 vs 94.5 | Helped |
| optimize expected return over the inferred environment distribution rather than using its MAP setting | robot-arm policy optimization with unknown p(theta) · Unknown p(theta), robotic arm | average cost: 28.76 -> 19.82 | Helped |
| select environment settings actively with BQ instead of sampling them randomly (ablation) | ALOQ · F-SRE1 and F-SRE2 | expected function value: F-SRE1: 0.081 vs 0.636; F-SRE2: 1.917 vs 2.407 | Helped |
| add Beta warping to the inputs (ablation) | ALOQ · Robotic arm, Collision Avoidance and Joint Breakage | expected cost: Collision Avoidance: 17.3 -> 8.9; Joint Breakage: 18.9 -> 7.7 | Helped |
| add intensification by re-evaluating selected policies (ablation) | ALOQ · Robotic arm, Collision Avoidance | expected cost: 74.1 -> 8.9 | Helped |
| add intensification by re-evaluating selected policies (ablation) | ALOQ · F-SRE1 and F-SRE2 | expected function value: F-SRE1: 0.646 vs 0.636; F-SRE2: 2.405 vs 2.407 | No clear effect |
| use one-step ALOQ without intensification (ablation) | ALOQ · Robotic arm tasks versus low-dimensional test functions | expected cost or function value: Collision Avoidance: 74.1 vs 8.9; Branin: 326.2 vs 330.0; Hartmann 6: 1.093 vs 0.937 | Mixed |
| use active BQ uncertainty reduction rather than uncertainty sampling (ablation) | ALOQ · Experiments with environment variables | expected return estimation | Helped |
| use ALOQ rather than WSN on problems with significant rare events | policy optimization · F-SRE1 and F-SRE2 | expected function value: F-SRE1: 0.636 vs 0.081; F-SRE2: 2.407 vs 1.917 | Helped |
Closest papers by meaning
- Distributionally Robust $Q$-Learning (ICML 2022 · similarity 0.90)
- Minimax Optimal Reinforcement Learning with Quasi-Optimism (ICLR 2025 · similarity 0.87)
- Bayesian Exploration Networks (ICML 2024 · similarity 0.87)
- Minimax Optimal and Computationally Efficient Algorithms for Distributionally Robust Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Augmented Bayesian Policy Search (ICLR 2024 · similarity 0.86)
- Bayesian Design Principles for Offline-to-Online Reinforcement Learning (ICML 2024 · similarity 0.86)
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.