Research map / Reinforcement learning / Visual and representation RL
Recurrent Predictive State Policy Networks
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 |
|---|---|---|---|
| Use RPSP-Alt instead of the strongest competing model | RPSP-Alt · Hopper | cumulative reward | Helped |
| Use an RPSP variant instead of the strongest competing model | RPSP · Swimmer | cumulative reward | Mixed |
| Use an RPSP variant instead of the strongest competing model | RPSP · Walker2d | cumulative reward | Mixed |
| Use an RPSP variant instead of the strongest competing model | RPSP · Cart-Pole | cumulative reward | Mixed |
| Use alternating optimization instead of joint VRPG optimization | RPSP · OpenAI Gym MuJoCo tasks | learning speed | Helped |
| Add two-stage regression initialization instead of random PSR initialization (ablation) | RPSP · Hopper | cumulative reward | Helped |
| Train the PSR parameters end-to-end instead of keeping them fixed (ablation) | RPSP · Hopper | cumulative reward | Helped |
| Add observation prediction loss regularization (ablation) | RPSP · Hopper | cumulative reward | Helped |
| Replace the PSR filter with a prediction-regularized GRU filter (ablation) | RPSP · Hopper | cumulative reward | Hurt |
| Concatenate a window of previous observations with the predictive state (ablation) | RPSP (+obs) · OpenAI Gym MuJoCo tasks | cumulative reward | Helped |
| Use RPSP instead of finite memory under mild observation noise | RPSP · Hopper | cumulative reward | Helped |
| Increase observation noise to excessive levels | RPSP versus finite memory · Hopper | cumulative reward | No clear effect |
Closest papers by meaning
- Efficient Recurrent Off-Policy RL Requires a Context-Encoder-Specific Learning Rate (NeurIPS 2024 · similarity 0.91)
- Bridging State and History Representations: Understanding Self-Predictive RL (ICLR 2024 · similarity 0.89)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.88)
- Recurrent Reinforcement Learning with Memoroids (NeurIPS 2024 · similarity 0.88)
- Real-Time Recurrent Reinforcement Learning (AAAI 2025 · similarity 0.88)
- Rethinking Transformers in Solving POMDPs (ICML 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.