Research map / Reinforcement learning / Visual and representation RL
SOLAR: Deep Structured Representations for Model-Based 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 |
|---|---|---|---|
| Replace SOLAR's representation and global models with a standard VAE (ablation) | SOLAR · Reacher | reward | Hurt |
| Replace SOLAR's representation and global models with a standard VAE (ablation) | SOLAR · Nonholonomic car | policy performance consistency across seeds | Hurt |
| Replace SOLAR's representation and global models with a standard VAE (ablation) | SOLAR · Sawyer Lego block stacking | task success | Hurt |
| Replace SOLAR's representation and global models with a standard VAE (ablation) | SOLAR · 2D navigation | task success | No clear effect |
| Fine-tune the model on newly collected data during the RL phase (ablation) | SOLAR · Experiments reported in this paper | final performance | No clear effect |
| Use PPO instead of SOLAR | SOLAR · 2D navigation | samples required to reach the compared performance: 1000 times | Mixed |
| Use PPO instead of SOLAR | SOLAR · Nonholonomic car | episodes required to learn a successful policy: over 25 times | Mixed |
| Use PPO instead of SOLAR | SOLAR · Reacher | episodes required to learn: about 40 times | Mixed |
| Pretrain and transfer a shared representation and global model across tasks | SOLAR · Sawyer Lego block stacking | total data requirements: about a factor of two | Helped |
| Use SOLAR instead of DVF | Sawyer pushing with sparse rewards · Sawyer pushing with sparse rewards | Final Distance to Goal (cm): 4.50±2.60 -> 1.85±0.86 | Helped |
Closest papers by meaning
- The Surprising Ineffectiveness of Pre-Trained Visual Representations for Model-Based Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
- Combining Model-Based and Model-Free Updates for Trajectory-Centric Reinforcement Learning (ICML 2017 · similarity 0.89)
- Towards General-Purpose Model-Free Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Simplifying Model-based RL: Learning Representations, Latent-space Models, and Policies with One Objective (ICLR 2023 · similarity 0.88)
- Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (AAAI 2025 · similarity 0.88)
- Video-Enhanced Offline Reinforcement Learning: A Model-Based Approach (ICML 2025 · similarity 0.87)
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.