Research map / Reinforcement learning / Policy gradient and actor-critic
Chunking the Critic: A Transformer-based Soft Actor-Critic with N-Step Returns
Dong Tian, Onur Celik, Gerhard Neumann
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 Transformer critic with N-step supervision to SAC | SAC · Box-Pushing dense | success rate (%): <=85 -> 96.8 | Improved |
| sparse-reward Box-Pushing, T-SAC vs TOP-ERL | TOP-ERL · Box-Pushing sparse | success rate (%): 70 -> 60 | Hurt |
| replace Transformer critic with multi-step MLP critic (ablation) | T-SAC · Box-Pushing dense | success rate | Hurt |
| remove causal mask from Transformer critic (ablation) | T-SAC · Box-Pushing dense | success rate | Improved |
| replace Transformer with GRU or LSTM critic (ablation) | T-SAC · Box-Pushing dense | success rate | Hurt |
| use fixed horizon instead of sampled N-step horizons (ablation) | T-SAC · Box-Pushing dense | success rate | Hurt |
| add multiple supervision windows per step (up to 4) (ablation) | T-SAC · Box-Pushing dense | success rate | Improved |
| add policy-side chunking to MLP critic (ablation) | MLP critic · Box-Pushing dense | success rate | No clear effect |
| add policy chunking length 4 to Transformer critic (ablation) | T-SAC · Box-Pushing dense | final success rate (%): 99.5 final success | Improved |
| increase critic freezing interval K to very large values (ablation) | T-SAC · Walker2d | return | Hurt |
| replace Polyak target with critic-freezing hard copy (ablation) | T-SAC · MuJoCo locomotion | return | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- TOP-ERL — TOP-ERL: Transformer-based Off-Policy Episodic Reinforcement Learning (ICLR 2025) · better on 0 of 1 shared result
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
- Generative Actor-Critic with Soft Bridge Policies (NeurIPS 2026 · similarity 0.88)
- Offline Actor-Critic Reinforcement Learning Scales to Large Models (ICML 2024 · similarity 0.88)
- Closing the Gap: Achieving Global Convergence (Last Iterate) of Actor-Critic under Markovian Sampling with Neural Network Parametrization (ICML 2024 · similarity 0.87)
- Transformers Provably Implement In-Context Reinforcement Learning with Policy Improvement (NeurIPS 2026 · similarity 0.87)
- ReLU to the Rescue: Improve Your On-Policy Actor-Critic with Positive Advantages (ICML 2024 · similarity 0.87)
- Langevin Soft Actor-Critic: Efficient Exploration through Uncertainty-Driven Critic Learning (ICLR 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.
