Research map / Reinforcement learning / Visual and representation RL
Learning a subspace of policies for online adaptation in Reinforcement Learning
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 |
|---|---|---|---|
| learn a line of policies in parameter space rather than a single policy | PPO/A2C policies · Brax HalfCheetah | average cumulative reward: 7697 -> 10589 | Helped |
| learn a line of policies in parameter space rather than a single policy | PPO/A2C policies · Brax Ant | average cumulative reward: 3338 -> 4031 | Helped |
| learn a line of policies in parameter space rather than a single policy | PPO/A2C policies · CartPole | average cumulative reward: 143.4 -> 149.9; DIAYN+R: 168.1 | Mixed |
| test more policies from the learned subspace during adaptation (K=5 to K=10) (ablation) | LoP · HalfCheetah test set | average cumulative reward: 10413 -> 10589 | Helped |
| test more policies from the learned subspace during adaptation (K=10 to K=20) (ablation) | LoP · HalfCheetah test set | average cumulative reward: 10589 -> 10691 | Helped |
| replace two-anchor line interpolation with a three-anchor Bézier curve (ablation) | LoP · HalfCheetah test set (K=10) | average cumulative reward: 10589 -> 8331 | Hurt |
| replace two-anchor line interpolation with a three-anchor convex combination (ablation) | LoP · HalfCheetah test set (K=10) | average cumulative reward: 10589 -> 9506 | Hurt |
| increase the DIAYN+R adaptation set from K=5 to K=10 (ablation) | DIAYN+R · HalfCheetah test set | average cumulative reward: 9114 -> 9680 | Helped |
Closest papers by meaning
- Provably Efficient Model-based Policy Adaptation (ICML 2020 · similarity 0.91)
- State Regularized Policy Optimization on Data with Dynamics Shift (NeurIPS 2023 · similarity 0.89)
- Efficient Test-Time Adaptation For Robot Policies (NeurIPS 2026 · similarity 0.89)
- OMPO: A Unified Framework for RL under Policy and Dynamics Shifts (ICML 2024 · similarity 0.88)
- The Ladder in Chaos: Improving Policy Learning by Harnessing the Parameter Evolving Path in A Low-dimensional Space (NeurIPS 2024 · similarity 0.88)
- Identifying Policy Gradient Subspaces (ICLR 2024 · similarity 0.88)
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.