Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Finding and Visualizing Weaknesses of Deep Reinforcement Learning Agents
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| replace a VAE reconstruction model with the full generator, including agent-perception and attentive reconstruction losses (ablation) | DQN inputs reconstructed by the generator · Pong | average score: -8 -> 14 | Helped |
| replace a VAE reconstruction model with the full generator, including agent-perception and attentive reconstruction losses (ablation) | DQN inputs reconstructed by the generator · Atlantis | average score: 95 -> 109 | Helped |
| replace a VAE reconstruction model with the full generator, including agent-perception and attentive reconstruction losses (ablation) | DQN inputs reconstructed by the generator · Q*bert | average score: 26 -> 31 | Helped |
| add agent-perception and KL losses to the reconstruction-only model (ablation) | DQN inputs reconstructed by the generator · Pong | average score: 4 -> 14 | Helped |
| add agent-perception and KL losses to the reconstruction-only model (ablation) | DQN inputs reconstructed by the generator · Atlantis | average score: 98 -> 109 | Helped |
| add agent-perception and KL losses to the reconstruction-only model (ablation) | DQN inputs reconstructed by the generator · Q*bert | average score: 28 -> 31 | Helped |
| replace real frames with full-generator reconstructions | DQN · Pong | average score: 14 -> 14 | No clear effect |
| replace real frames with full-generator reconstructions | DQN · Atlantis | average score: 108 -> 109 | Helped |
| replace real frames with full-generator reconstructions | DQN · Q*bert | average score: 64 -> 31 | Hurt |
Closest papers by meaning
- Efficient Reinforcement Learning Through Adaptively Pretrained Visual Encoder (AAAI 2025 · similarity 0.86)
- Sliding Puzzles Gym: A Scalable Benchmark for State Representation in Visual Reinforcement Learning (ICML 2025 · similarity 0.86)
- Attention Trajectories as a Diagnostic Axis for Deep Reinforcement Learning (NeurIPS 2026 · similarity 0.86)
- Investigating Pre-Training Objectives for Generalization in Vision-Based Reinforcement Learning (ICML 2024 · similarity 0.86)
- DrM: Mastering Visual Reinforcement Learning through Dormant Ratio Minimization (ICLR 2024 · similarity 0.86)
- Towards Principled Representation Learning from Videos for Reinforcement Learning (ICLR 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.