Research map / Reinforcement learning / Imitation learning
APC-RL: Exceeding data-driven behavior priors with adaptive policy composition
Finn Rietz, Pedro Zuidberg Dos Martires, Johannes A. Stork
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 |
|---|---|---|---|
| use arbitrator selector instead of learned hierarchical selector (ablation) | APC · CarRacing | return | Improved |
| add reward sharing across actors (ablation) | APC · CarRacing | return | Improved |
| add prior-free actor with APC composition vs PARROT under misaligned priors | APC · PointMaze (misaligned) | success rate: 0% -> 100% | Improved |
| compose prior-based and prior-free actors instead of from-scratch SAC | APC · PointMaze (misaligned) | time to success | Improved |
| use APC instead of PARROT with aligned priors | APC · PointMaze/FrankaKitchen (aligned) | convergence speed | Hurt |
| use APC instead of SAC on CarRacing with human demonstrations | APC · CarRacing | steps to optimal return: 250k -> 30k | Improved |
| use APC instead of QFilter on CarRacing | APC · CarRacing | steps to optimal return: 100k -> 30k | Improved |
| increase the number of misaligned priors alongside an aligned prior (ablation) | APC · PointMaze | time to success | Hurt |
| provide fully misaligned priors only | APC · PointMaze | time to success | Hurt |
| raise arbitrator temperature to 100 (ablation) | APC · PointMaze (misaligned prior) | running success rate | Hurt |
| vary arbitrator temperature from 0.01 to 10 (ablation) | APC · PointMaze | running success rate | No clear effect |
Closest papers by meaning
- A Snapshot of Influence: A Local Data Attribution Framework for Online Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Composite Flow Matching for Reinforcement Learning with Shifted-Dynamics Data (NeurIPS 2025 · similarity 0.87)
- Demonstration-Regularized RL (ICLR 2024 · similarity 0.87)
- Hybrid Inverse Reinforcement Learning (ICML 2024 · similarity 0.86)
- Risk-Conditioned Reinforcement Learning: A Generalized Approach for Adapting to Varying Risk Measures (AAAI 2024 · similarity 0.86)
- Reinforcement Learning with Adaptive Regularization for Safe Control of Critical Systems (NeurIPS 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.
