Research map / Reinforcement learning / Visual and representation RL
MoDem: Accelerating Visual Model-Based Reinforcement Learning with Demonstrations
Nicklas Hansen, Yixin Lin, Hao Su, Xiaolong Wang, Vikash Kumar, Aravind Rajeswaran
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 MoDem's three-phase demonstration framework with TD-MPC | TD-MPC on Adroit, Meta-World, and DMControl · 21 visual control tasks | success rate or episode return: 160%-250% more successful | Improved |
| Add policy pretraining by behavior cloning (ablation) | MoDem · Adroit | success rate | Improved |
| Seed world-model learning with rollouts from the pretrained BC policy rather than random interactions (ablation) | MoDem · Adroit | success rate | Improved |
| Oversample demonstrations during interactive learning instead of merely appending them to the buffer (ablation) | MoDem · Adroit | success rate | Improved |
| Append demonstrations to TD-MPC's replay buffer without MoDem's targeted phases and oversampling (ablation) | TD-MPC · Adroit | success rate | No clear effect |
| Increase the number of demonstrations (ablation) | MoDem · Adroit | success rate | Improved |
| Use a single demonstration instead of five (ablation) | MoDem · Adroit | success rate | Mixed |
| Vary how much interaction data is collected using the pretrained policy (ablation) | MoDem · Adroit | success rate | No clear effect |
| Add random image shift augmentation to end-to-end representation learning (ablation) | MoDem · Adroit | success rate | Improved |
| Anneal demonstration sampling from 75% to 25% instead of keeping it at 25% (ablation) | MoDem · Adroit | success rate | No clear effect |
| Use oracle-policy rather than human demonstrations (ablation) | MoDem · Adroit | success rate | Mixed |
| Learn the visual representation end-to-end rather than using frozen R3M features (ablation) | MoDem · Adroit | success rate | Improved |
Later papers that compared with this method
3 later accepted papers report MoDem as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- SLOPE: Optimistic Potential Landscape Shaping for Model-based Reinforcement Learning (NeurIPS 2026 · better on 6 of 6 shared results)
- Uncertainty-Guided Exploration and Stable Planning for Sparse-Reward Manipulation from Limited Demonstrations (ICML 2026 · better on 2 of 2 shared results)
- Multi-Stage Manipulation with Demonstration-Augmented Reward, Policy, and World Model Learning (ICML 2025 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards General-Purpose Model-Free Reinforcement Learning (ICLR 2025 · similarity 0.90)
- DrM: Mastering Visual Reinforcement Learning through Dormant Ratio Minimization (ICLR 2024 · similarity 0.89)
- The Surprising Ineffectiveness of Pre-Trained Visual Representations for Model-Based Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
- Overcoming Slow Decision Frequencies in Continuous Control: Model-Based Sequence Reinforcement Learning for Model-Free Control (ICLR 2025 · similarity 0.89)
- PWM: Policy Learning with Multi-Task World Models (ICLR 2025 · similarity 0.89)
- Video-Enhanced Offline Reinforcement Learning: A Model-Based Approach (ICML 2025 · 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.
