Research map / Reinforcement learning / Visual and representation RL
Simplifying Model-based RL: Learning Representations, Latent-space Models, and Policies with One Objective
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 |
|---|---|---|---|
| jointly train the representation, latent model, and policy with the ALM objective instead of using separate objectives | ALM · locomotion benchmark | return | Helped |
| use ALM without ensembles | model-based RL on continuous-control tasks · locomotion tasks | wall-clock training time: ∼6× less wall-clock time | Helped |
| use ALM instead of SAC | continuous-control RL · near-optimal return training | wall-clock time: about 50\% less time | Helped |
| remove the KL consistency term from the encoder objective (ablation) | ALM · continuous-control tasks | return | Hurt |
| add the KL consistency term to the encoder objective (ablation) | ALM · continuous-control tasks | return | Helped |
| remove the classifier-based reward term from the policy objective (ablation) | ALM · Ant and Walker | learning stability | Hurt |
| add the classifier-based reward term to the policy objective (ablation) | ALM · Ant and Walker | learning stability | Helped |
| remove the value term from the encoder objective (ablation) | ALM · continuous-control tasks | return | No clear effect |
| vary the number of latent-model unrolling steps K (ablation) | ALM · continuous-control tasks | return | Mixed |
| use ALM(3)-learned representations with a TD3 actor loss (ablation) | TD3 · continuous-control tasks | return | Helped |
| use the logarithm-transformed reward with a=10000 (ablation) | ALM · continuous-control tasks | return | No clear effect |
| set classifier-reward coefficient c to 0.1 instead of its theoretical value 1 (ablation) | ALM · continuous-control tasks | return | Helped |
Closest papers by meaning
- Mismatched No More: Joint Model-Policy Optimization for Model-Based RL (NeurIPS 2022 · similarity 0.90)
- Learning to Reuse Policies in State Evolvable Environments (ICML 2025 · similarity 0.88)
- Towards General-Purpose Model-Free Reinforcement Learning (ICLR 2025 · similarity 0.88)
- Model-based RL as a Minimalist Approach to Horizon-Free and Second-Order Bounds (ICLR 2025 · similarity 0.87)
- Offline RL with Observation Histories: Analyzing and Improving Sample Complexity (ICLR 2024 · similarity 0.87)
- Are Expressive Models Truly Necessary for Offline RL? (AAAI 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.