Research map / Reinforcement learning / Visual and representation RL
Actor-Critic for Continuous Action Chunks: A Reinforcement Learning Framework for Long-Horizon Robotic Manipulation with Sparse Reward
Jiarui Yang, Bin Zhu, Jingjing Chen, Yu-Gang Jiang
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 |
|---|---|---|---|
| remove asymmetric update, let actor learn from all trajectories (ablation) | AC3 · BiGym move plate | success rate | Improved |
| add intrinsic anchor reward r_int=0.1 vs no reward shaping (ablation) | AC3 · BiGym move plate | success rate | Improved |
| add intrinsic anchor reward r_int=0.1 vs no reward shaping (ablation) | AC3 · RLBench open oven | success rate | No clear effect |
| set anchor reward equal to success reward or linear in trajectory length (ablation) | AC3 · BiGym move plate | success rate | Hurt |
| chunk size C=4 instead of 16 (ablation) | AC3 · BiGym move plate | success rate | Hurt |
| longer chunk size beyond moderate (ablation) | AC3 · RLBench open oven | success rate | Hurt |
| n-step return n=C=16 instead of n=4 (ablation) | AC3 · BiGym move plate | success rate | Hurt |
| n-step return n=C=16 instead of n=4 (ablation) | AC3 · RLBench open oven | success rate | Improved |
| n=1 instead of n=4 or 8 (ablation) | AC3 · higher-dimensional tasks | success rate | Hurt |
| AC3 vs chunk-wise BC | AC3 · RLBench (10 tasks) | success rate | Improved |
| AC3 vs CQN-AS | AC3 · RLBench (10 tasks) | success rate | No clear effect |
| per-step DrQ-v2 vs chunk-wise BC | DrQ-v2 · RLBench (10 tasks) | success rate | Hurt |
Closest papers by meaning
- Multi-Stage Manipulation with Demonstration-Augmented Reward, Policy, and World Model Learning (ICML 2025 · similarity 0.90)
- Reinforcement Learning with Action Chunking (NeurIPS 2025 · similarity 0.88)
- Learning Process Rewards via Visitation Matching for Efficient RL (NeurIPS 2026 · similarity 0.88)
- Coarse-to-fine Q-Network with Action Sequence for Data-Efficient Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- PACE: Phase-Aware Chunk Execution for Robot Policies with Action Chunking (NeurIPS 2026 · similarity 0.87)
- Offline Actor-Critic Reinforcement Learning Scales to Large Models (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.
