Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Inherently Explainable Reinforcement Learning in Natural Language
XIANGYU PENG, Mark Riedl, Prithviraj Ammanabrolu
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 HEX-RL hierarchical graph attention agent instead of baselines | HEX-RL (game and IM) · Jericho 9 games | % completion (max score): 34.9 -> 35.8 | Improved |
| add intrinsic motivation reward to HEX-RL (ablation) | HEX-RL · Jericho 9 games | % completion (final score): 27.2 -> 28.2 | Improved |
| compare HEX-RL game and IM against Q*BERT on final score | HEX-RL · Jericho 9 games | % completion (final score): 30.8 -> 28.2 | Hurt |
| use Hierarchical Graph Attention explanations instead of LSTM attention | HEX-RL · zork1, library, balances | human preference (confidence, understandability) | Improved |
| use Hierarchical Graph Attention explanations instead of LSTM attention | HEX-RL · zork1, library, balances | human-likeness preference | Mixed |
| use temporal instead of immediate explanations | HEX-RL · human judgment | goal context | Improved |
| use temporal instead of immediate explanations | HEX-RL · human judgment | coherence and readability | No clear effect |
| add CALM language model filter to Bayes filter (ablation) | temporal explanations · human evaluation | confidence and understandability | Improved |
| add semantic filter to Bayes+CALM (ablation) | temporal explanations · human evaluation | confidence, human-likeness, understandability | Improved |
| use HEX-RL with intrinsic motivation instead of Q*BERT | HEX-RL · pentari | average score: 50 -> 44.7 | Hurt |
| use HEX-RL with intrinsic motivation instead of Q*BERT | HEX-RL · detective | average score: 274 -> 276.9 | No clear effect |
Closest papers by meaning
- CausalXRL: Explainable Reinforcement Learning through Causal Graph Reasoning (ICML 2026 · similarity 0.89)
- Deep Reinforcement Learning with Stacked Hierarchical Attention for Text-based Games (NeurIPS 2020 · similarity 0.89)
- Tell me why! Explanations support learning relational and causal structure (ICML 2022 · similarity 0.89)
- Explaining Reinforcement Learning Agents through Counterfactual Action Outcomes (AAAI 2024 · similarity 0.87)
- ExPO: Unlocking Hard Reasoning with Self-Explanation-Guided Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- RLZero: Direct Policy Inference from Language Without In-Domain Supervision (NeurIPS 2025 · 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.
