Research map / Adversarial robustness and security / Attacks on text and graphs
Robust Deep Reinforcement Learning against Adversarial Behavior Manipulation
Shojiro Yamabe, Kazuto Fukuchi, Jun Sakuma
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 time-discounting (gamma^t) to the smoothness regularizer (ablation) | SA-PPO policy smoothing · Meta-World (10 tasks) | clean reward: avg +28.2%; drawer-close 2156 -> 4237 | Improved |
| add time-discounting to policy smoothing (ablation) | SA-PPO · Meta-World drawer-close | best attack reward: 4 -> 4860 | Hurt |
| add time-discounting to policy smoothing (ablation) | SA-PPO · Meta-World window-close | best attack reward: 485 -> 482 | No clear effect |
| add time-discounting to policy smoothing (ablation) | SA-PPO · Meta-World handle-pull-side | best attack reward: 10 -> 7 | No clear effect |
| add time-discounting to policy smoothing (ablation) | SA-PPO · Meta-World faucet-close | best attack reward: 1559 -> 1789 | Hurt |
| add policy smoothing (SA-PPO or TDRT) to PPO | PPO · Meta-World window-close | best attack reward: 4505 -> 482 | Improved |
| adversarial training (ATLA-PPO, PA-ATLA-PPO) added to PPO | PPO · Meta-World window-close | best attack reward: 4505 -> 4270 / 4041 | No clear effect |
| use imitation-learning attack BIA instead of targeted PGD | adversary · Meta-World drawer-close | attack reward: 2891 -> 4760 | Improved |
| remove victim action access (ILfO no-box vs ILfD black-box) | BIA · Meta-World faucet-open | attack reward: 3031 -> 2718 | Hurt |
| BIA (demonstrations only) vs reward-maximization SA-RL | BIA · Meta-World faucet-open | attack reward: 1448 -> 3031 | Improved |
| BIA vs SA-RL | BIA · Meta-World window-close | attack reward: 4505 -> 3962 | Hurt |
Closest papers by meaning
- RAT: Adversarial Attacks on Deep Reinforcement Agents for Targeted Behaviors (AAAI 2025 · similarity 0.91)
- Representation-Based Robustness in Goal-Conditioned Reinforcement Learning (AAAI 2024 · similarity 0.88)
- Optimal Attack and Defense for Reinforcement Learning (AAAI 2024 · similarity 0.88)
- Rethinking Adversarial Policies: A Generalized Attack Formulation and Provable Defense in RL (ICLR 2024 · similarity 0.88)
- Diffusion Guided Adversarial State Perturbations in Reinforcement Learning (NeurIPS 2025 · similarity 0.87)
- Diffusion Policy Attacker: Crafting Adversarial Attacks for Diffusion-based Policies (NeurIPS 2024 · 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.
