Research map / Reinforcement learning / Visual and representation RL
EDGE: Explaining Deep Reinforcement Learning Policies
Wenbo Guo, Xian Wu, Usmann Khan, Xinyu Xing
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 EDGE explanations instead of best baseline to pick attack steps | explanation-guided DRL agent intervention · Pong attack | win rate change (points): -30.33 -> -65.47 | Improved |
| use EDGE explanations instead of Rudder to pick attack steps | explanation-guided DRL agent intervention · You-Shall-Not-Pass attack | win rate change (points): -32.53 -> -35.13 | No clear effect |
| use EDGE explanations instead of Attention RNN for attack | explanation-guided DRL agent intervention · Kick-And-Defend attack | win rate change (points): -41.20 -> -43.47 | No clear effect |
| use EDGE explanations for policy patching vs Rationale Net | explanation-guided DRL agent intervention · Pong patching | win rate change (points): -3.66 -> +2.75 | Improved |
| use EDGE explanations for policy patching vs Rudder | explanation-guided DRL agent intervention · You-Shall-Not-Pass patching | win rate change (points): +1.76 -> +2.91 | Improved |
| use EDGE explanations for policy patching vs Saliency | explanation-guided DRL agent intervention · Kick-And-Defend patching | win rate change (points): +1.17 -> +1.21 | No clear effect |
| blind victim at EDGE-identified critical steps vs Rudder | explanation-guided DRL agent intervention · You-Shall-Not-Pass robustifying | victim win rate change (points): +8.54 -> +35.30 | Improved |
| EDGE explanation vs Attention/Rationale Net/saliency/Rudder | explanation-guided DRL agent intervention · Pong, MuJoCo-Y, MuJoCo-K | explanation fidelity score | Improved |
| self-explainable Attention and Rationale Net vs post-training explanations | explanation-guided DRL agent intervention · Pong, MuJoCo | explanation fidelity score | No clear effect |
Later papers that compared with this method
2 later accepted papers report EDGE as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Duolando: Follower GPT with Off-Policy Reinforcement Learning for Dance Accompaniment (ICLR 2024 · better on 1 of 1 shared result)
- Learning to Identify Critical States for Reinforcement Learning from Videos (ICCV 2023 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Rudder — RUDDER: Return Decomposition for Delayed Rewards (NeurIPS 2019) · better on 7 of 7 shared results
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
- Local Explanations for Reinforcement Learning (AAAI 2023 · similarity 0.89)
- Explaining Reinforcement Learning Agents through Counterfactual Action Outcomes (AAAI 2024 · similarity 0.87)
- CrystalBox: Future-Based Explanations for Input-Driven Deep RL Systems (AAAI 2024 · similarity 0.87)
- RICE: Breaking Through the Training Bottlenecks of Reinforcement Learning with Explanation (ICML 2024 · similarity 0.87)
- ORCAID: Oblique Rule-Based Continuous-Action Interpretation for Deep RL Policies (NeurIPS 2026 · similarity 0.87)
- Approximating Shapley Explanations in Reinforcement Learning (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.
