Research map / Reinforcement learning / Visual and representation RL
TaskLoom: Weaving Knowledge Across Tasks in World Models
Qingzhang Zeng, Peixi Peng, Hang Li, Luntong Li, Yonghong Tian
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 |
|---|---|---|---|
| add gradient-based task grouping (one world model per group) instead of one shared world model for all tasks (ablation) | Shared-Dreamer · Proprio Control | mean return: 280.3 -> 556.8 | Improved |
| naively train all tasks jointly in a single shared world model (ablation) | DreamerV3 · Proprio Control | mean return: 456.1 -> 280.3 | Hurt |
| add inter-group shared knowledge (SK) base in Stage 2 (ablation) | G-Baseline · Proprio Control | mean return: 556.8 -> 647.4 | Improved |
| add inter-group shared knowledge (SK) base in Stage 2 (ablation) | G-Baseline · Proprio Control | median return: 533.4 -> 775.7 | Improved |
| keep group-specific dynamics predictor decoupled from the shared knowledge base instead of absorbing it (ablation) | TaskLoom · Proprio Control | mean return: 604.5 -> 647.4 | Improved |
| keep group-specific dynamics predictor decoupled from the shared knowledge base instead of absorbing it (ablation) | TaskLoom · Proprio Control | median return: 704.7 -> 775.7 | Improved |
| add task grouping without cross-group knowledge flow (ablation) | Shared-Dreamer · Hopper Hop (Proprio) | return: 3.9 -> 0.6 | Hurt |
| apply TaskLoom cross-task knowledge sharing instead of single-task DreamerV3 | DreamerV3 · Visual Control Reacher Hard | return: 53.8 -> 466.0 | Improved |
| apply TaskLoom instead of single-task DreamerV3 | DreamerV3 · Proprio Control | mean return: 456.1 -> 647.4 | Improved |
| apply TaskLoom instead of single-task DreamerV3 | DreamerV3 · Visual Control | mean return: 382.0 -> 455.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CURL — CURL: Contrastive Unsupervised Representations for Reinforcement Learning (ICML 2020) · better on 2 of 2 shared results
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
- Reinforcing VLAs in Task-Agnostic World Models (NeurIPS 2026 · similarity 0.89)
- PWM: Policy Learning with Multi-Task World Models (ICLR 2025 · similarity 0.88)
- RLVR-World: Training World Models with Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- Scaling Offline Model-Based RL via Jointly-Optimized World-Action Model Pretraining (ICLR 2025 · similarity 0.87)
- Making Offline RL Online: Collaborative World Models for Offline Visual Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- HarmonyDream: Task Harmonization Inside World Models (ICML 2024 · 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.
