Research map / Continual learning / Online and replay-based
COOM: A Game Benchmark for Continual Reinforcement Learning
Tristan Tomilin, Meng Fang, Yudi Zhang, Mykola Pechenizkiy
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 |
|---|---|---|---|
| make tasks harder (COC instead of CO8) | PackNet · COOM | average performance: 0.82 -> 0.20 | Hurt |
| make tasks harder (COC instead of CO8) | MAS · COOM | average performance: 0.58 -> 0.04 | Hurt |
| lengthen cross-domain sequence from 4 to 8 tasks | MAS · COOM cross-domain | average performance: 0.55 -> 0.82 | Improved |
| lengthen cross-domain sequence from 4 to 8 tasks | L2 · COOM cross-domain | average performance: 0.71 -> 0.87 | Improved |
| lengthen cross-objective sequence from 4 to 8 tasks | L2 · COOM cross-objective | average performance: 0.80 -> 0.71 | Hurt |
| lengthen cross-objective sequence from 4 to 8 tasks | MAS · COOM cross-objective | average performance: 0.72 -> 0.58 | Hurt |
| prepend first-half tasks (CO82nd vs CO4) | all methods · COOM cross-objective | average performance: 0.618 -> 0.643 | Improved |
| prepend first-half tasks (CD82nd vs CD4) | all methods · COOM cross-domain | average performance: 0.595 -> 0.673 | Improved |
| use AGEM instead of naive fine-tuning | SAC · COOM average | average performance: 0.40 -> 0.28 | Hurt |
| use VCL instead of naive fine-tuning | SAC · COOM average | average performance: 0.40 -> 0.33 | Hurt |
| use PackNet instead of fine-tuning | SAC · COOM average | forgetting: 0.53 -> 0.01 | Improved |
| weaken regularization strength in L2/EWC | L2 and EWC · COOM | stability and plasticity | No clear effect |
Closest papers by meaning
- C-NAV: Towards Self-Evolving Continual Object Navigation in Open World (NeurIPS 2025 · similarity 0.86)
- HASARD: A Benchmark for Vision-Based Safe Reinforcement Learning in Embodied Agents (ICLR 2025 · similarity 0.86)
- Overcoming Catastrophic Forgetting in Visual Continual Learning with Reinforcement Fine-Tuning (NeurIPS 2026 · similarity 0.86)
- Online Continual Learning for Interactive Instruction Following Agents (ICLR 2024 · similarity 0.86)
- When Should We Prefer State-to-Visual DAgger over Visual Reinforcement Learning? (AAAI 2025 · similarity 0.86)
- Continual Reinforcement Learning by Planning with Online World Models (ICML 2025 · similarity 0.85)
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.
