Research map / Multi-agent systems and games / Games and equilibria
Playing FPS Games with Deep Reinforcement Learning
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 |
|---|---|---|---|
| Add a navigation network to the action network (ablation) | Arnold · Limited Deathmatch, known map | Kill to Death Ratio: 4.64 -> 5.52 | Helped |
| Add a navigation network to the action network (ablation) | Arnold · Full Deathmatch, train maps | Kill to Death Ratio: 2.83 -> 4.58 | Helped |
| Add a navigation network to the action network (ablation) | Arnold · Full Deathmatch, test maps | Kill to Death Ratio: 3.12 -> 6.94 | Helped |
| Co-train the DRQN with enemy-presence game features (ablation) | Action network · Limited Deathmatch | K/D score: less than 2.0 -> over 4.0 | Helped |
| Add dropout (ablation) | Action network · Limited Deathmatch | Enemy detection accuracy: 70% -> 90% | Helped |
| Use game features as input rather than co-training targets (ablation) | DQN · Deathmatch | performance | No clear effect |
| Use a separate feature-prediction network and reinject predictions into the LSTM (ablation) | DRQN · Deathmatch | performance | No clear effect |
| Use grayscale images instead of color (ablation) | Arnold · Deathmatch | performance | Hurt |
| Use a DQN with multiple stacked frames for game-feature detection (ablation) | DQN · Deathmatch | enemy detection accuracy: 70% | Hurt |
| Increase frame skip beyond 4 (ablation) | Arnold · Deathmatch | performance | Hurt |
| Train a single network for the whole deathmatch task instead of phase-specific networks (ablation) | Arnold · Deathmatch | performance | Hurt |
| Add a small positive reward proportional to distance travelled (ablation) | Navigation network · Navigation | exploration speed | Helped |
Closest papers by meaning
- Combo-Action: Training Agent For FPS Game with Auxiliary Tasks (AAAI 2019 · similarity 0.91)
- Training Agent for First-Person Shooter Game with Actor-Critic Curriculum Learning (ICLR 2017 · similarity 0.88)
- Vision-Zero: Scalable VLM Self-Improvement via Strategic Gamified Self-Play (ICLR 2026 · similarity 0.86)
- Investigating Pre-Training Objectives for Generalization in Vision-Based Reinforcement Learning (ICML 2024 · similarity 0.86)
- Diffusion Models Are Real-Time Game Engines (ICLR 2025 · similarity 0.85)
- Towards Building Human-like Smart Agents in Modern 3D Video Games (Student Abstract) (AAAI 2025 · similarity 0.85)
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.