Research map / Reinforcement learning / Policy gradient and actor-critic
Stabilizing Dynamical Systems via Policy Gradient Methods
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 |
|---|---|---|---|
| Train discount annealing with larger initial-state radius r (ablation) | Discount annealing · Nonlinear cart-pole | ROA radius: r=0.1: [0.702, 0.704]; r=0.7: [0.769, 0.777] | Helped |
| Optimize policy on the nonlinear dynamics rather than only the linearization | Discount annealing · Nonlinear cart-pole; final discount γ=1 | cost: [16.5, 16.8] -> [15.2, 15.4] | Helped |
| Optimize policy on the nonlinear dynamics rather than only the linearization | Discount annealing · Nonlinear cart-pole; final discount γ=1 | ROA radius: [0.702, 0.703] -> [0.769, 0.777] | Helped |
| Use training radius r>0.1 instead of the linearization-only baselines (ablation) | Discount annealing · Nonlinear cart-pole | ROA radius: Discount annealing: [0.711, 0.713], [0.727, 0.734], [0.731, 0.744], [0.769, 0.777]; LQR: 0.703; H-infinity: 0.506 | Helped |
| Increase training radius to approximately r=0.75 (ablation) | Discount annealing · Nonlinear cart-pole | successful stabilization: r~\le~ 0.7; r\approx 0.75 | Hurt |
| Increase training radius from r=0.05 to r=0.7 (ablation) | Policy-gradient controller · Nonlinear cart-pole | controller difference from linearized LQR: r=0.05: K_pg^⋆(γ_t) ≈ K_lin^⋆(γ_t); r=0.7: difference steadily increases | Mixed |
Closest papers by meaning
- Learning Control-Oriented Dynamical Structure from Data (ICML 2023 · similarity 0.88)
- The Power of Learned Locally Linear Models for Nonlinear Policy Optimization (ICML 2023 · similarity 0.88)
- Performative Control for Linear Dynamical Systems (NeurIPS 2024 · similarity 0.88)
- On the Sample Complexity of Stabilizing LTI Systems on a Single Trajectory (NeurIPS 2022 · similarity 0.87)
- Online Policy Gradient for Model Free Learning of Linear Quadratic Regulators with $\sqrt$T Regret (ICML 2021 · similarity 0.87)
- A Memory Efficient Unified Algorithm for Online Learning of Linear Dynamical Systems (NeurIPS 2026 · 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.