Research map / Continual learning / Online and replay-based
Continual World: A Robotic Benchmark For Continual 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 |
|---|---|---|---|
| Use PopArt reward normalization for multi-task training (ablation) | MT · CW20 | performance: 0.51 -> 0.65 | Helped |
| Regularize the critic network (ablation) | EWC · CW20 | performance | Hurt |
| Do not reset the replay buffer at task change (ablation) | SAC continual-learning agents · Diagonal task transfer | forward transfer | Helped |
| Use the current policy rather than uniform actions for exploration after task change (ablation) | SAC continual-learning agents · Diagonal task transfer | forward transfer | Helped |
| Increase replay batch size to 512 (ablation) | Perfect Memory · CW20 | performance | No clear effect |
| Use 1 weight sample instead of 10 for VCL prediction averaging (ablation) | VCL · CW20 | performance | No clear effect |
| Apply a continual-learning method to the critic (ablation) | SAC continual-learning agents · CW20 | performance | Hurt |
Closest papers by meaning
- Continual Reinforcement Learning by Planning with Online World Models (ICML 2025 · similarity 0.91)
- Continual Learning Bench: Evaluating Frontier AI Systems in Real-World Stateful Environments (NeurIPS 2026 · similarity 0.88)
- Knowledge Retention in Continual Model-Based Reinforcement Learning (ICML 2025 · similarity 0.88)
- On-Policy Algorithms for Continual Reinforcement Learning (Student Abstract) (AAAI 2025 · similarity 0.88)
- Online Continual Learning for Interactive Instruction Following Agents (ICLR 2024 · similarity 0.87)
- Position: Lifetime tuning is incompatible with continual reinforcement learning (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.