Research map / Continual learning / Online and replay-based
Resolving the Stability-Plasticity Dilemma in Reinforcement Learning via Complementary Continual Critics
Bo Sun, Peixi Peng, Guang Tan, Haoran Xu, Yaokun Li, Yiqian Chang, Shuaixian Wang, Luntong Li
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 |
|---|---|---|---|
| apply CBP to one critic only or EWC to one critic only (ablation) | CD-CCA · Quadruped Walk and CARLA Highway | training return | Improved |
| replace cross-attention fusion with minimum of two critics (ablation) | CD-CCA · Quadruped Walk and CARLA Highway | training return | Hurt |
| use cross-attention fusion without CBP and EWC (ablation) | SAC baseline · Quadruped Walk and CARLA Highway | sample efficiency | Improved |
| use two CBP critics instead of CBP plus EWC (ablation) | CD-CCA · Quadruped Walk and CARLA Highway | training return | Hurt |
| use two EWC critics instead of CBP plus EWC (ablation) | CD-CCA · Quadruped Walk and CARLA Highway | training return | Hurt |
| extend SAC and TD3 to four critics (two CBP, two EWC) (ablation) | SAC, TD3 · Quadruped Walk and CARLA Highway | convergence | Improved |
| add CD-CCA to DrQ-v2 | DrQ-v2 · DMControl 4 hard tasks (500k) | average episode return: 636.8 -> 683.0 | Improved |
| add CD-CCA to DrQ-v2 | DrQ-v2 · DMControl Pendulum Swingup (500k) | episode return: 842±12 -> 756±119 | Hurt |
| add CD-CCA to DrQ-v2 at 1M steps | DrQ-v2 · DMControl Pendulum Swingup (1M) | episode return: 812±23 -> 817±17 | No clear effect |
| add CD-CCA to DeepMDP | DeepMDP · CARLA Highway | episode reward: 170±36 -> 343±63 | Improved |
| add CD-CCA to DeepMDP | DeepMDP · CARLA Highway | driving distance (m): 132±20 -> 287±45 (ResAct 299±24 still higher) | Mixed |
Closest papers by meaning
- Revisiting Plasticity in Visual Reinforcement Learning: Data, Modules and Training Stages (ICLR 2024 · similarity 0.89)
- Overcoming Catastrophic Forgetting in Visual Continual Learning with Reinforcement Fine-Tuning (NeurIPS 2026 · similarity 0.88)
- A Study of Plasticity Loss in On-Policy Deep Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Rethinking the Stability-Plasticity Trade-off in Continual Learning from an Architectural Perspective (ICML 2025 · similarity 0.86)
- Generalizing Consistency Policy to Visual RL with Prioritized Proximal Experience Regularization (NeurIPS 2024 · similarity 0.86)
- Self-Composing Policies for Scalable Continual Reinforcement Learning (ICML 2024 · 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.
