Research map / Reinforcement learning / Offline RL
Learning Massively Multitask World Models for Continuous Control
Nicklas Hansen, Hao Su, Xiaolong Wang
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 |
|---|---|---|---|
| Condition the agent on language instructions (ablation) | Newt · MMBench 200 tasks | normalized score: 0.371 -> 0.438 | Improved |
| Pretrain on demonstrations then finetune online instead of training from scratch | Newt · MMBench transfer tasks, zero-shot | normalized score: 0.013 -> 0.192 | Improved |
| Pretrain on demonstrations then finetune online instead of training from scratch | Newt · MMBench transfer tasks, 100k steps | normalized score: 0.480 -> 0.868 | Improved |
| Add RGB image observations to state agent (ablation) | Newt · MMBench all domains | normalized score: 0.438 -> 0.442 | No clear effect |
| Add RGB image observations (ablation) | Newt · MMBench RoboDesk | normalized score: +0.125 (to 0.500) | Improved |
| Add RGB image observations (ablation) | Newt · MMBench DMControl | normalized score: -0.029 (to 0.471) | Hurt |
| Replace unseen object noun with 'cube' in instruction (ablation) | Newt · Unseen push tasks | success rate: 0.3 -> 21.0 | Improved |
| Replace unseen object noun with 'cube' in instruction (ablation) | Newt · Unseen pick-and-place tasks | success rate: 10.5 -> 0 | Hurt |
| Use any single way of using demonstrations (pretraining, oversampling, BC loss) (ablation) | Newt · MMBench | normalized score | Improved |
| Scale model and batch size (ablation) | Newt · MMBench 200 tasks | normalized score | Improved |
| Use Newt instead of PPO, FastTD3 and TD-MPC2 | Newt · MMBench state obs | normalized score | Improved |
| Use Newt on MuJoCo, Box2D and Atari domains vs BC | Newt · MMBench MuJoCo/Box2D/Atari | normalized score | No clear effect |
Closest papers by meaning
- PWM: Policy Learning with Multi-Task World Models (ICLR 2025 · similarity 0.88)
- Continual Reinforcement Learning by Planning with Online World Models (ICML 2025 · similarity 0.87)
- TD-MPC2: Scalable, Robust World Models for Continuous Control (ICLR 2024 · similarity 0.87)
- Representation Learning Enables Scalable Multitask Deep Reinforcement Learning (NeurIPS 2026 · similarity 0.87)
- XLand-100B: A Large-Scale Multi-Task Dataset for In-Context Reinforcement Learning (ICLR 2025 · similarity 0.86)
- Overcoming Slow Decision Frequencies in Continuous Control: Model-Based Sequence Reinforcement Learning for Model-Free Control (ICLR 2025 · similarity 0.86)
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.
