Researchers have developed the Dark Souls Learning Environment (DSLE), a platform designed to benchmark AI agents against the boss encounters in Dark Souls: Remastered. The environment presents 22 boss fights as challenges with high-dimensional visual input and sparse rewards, requiring agents to perform real-time actions. Initial evaluations on a five-boss subset (DSLE-5) showed that even advanced agents like Proximal Policy Optimization (PPO) and Deep Q-Network (DQN) failed to demonstrate significant learning, with win rates below 1% on the tutorial boss and 0% on others, despite extensive training. AI
IMPACT This benchmark highlights the significant challenges in developing AI agents capable of complex, real-time combat with sparse rewards, indicating a need for more advanced reinforcement learning techniques.
RANK_REASON The cluster contains a research paper detailing a new benchmark environment for AI agents.
Read on arXiv cs.NE (Neural & Evolutionary) →
- arXiv
- Asylum Demon
- Dark Souls: Remastered
- Deep Q-Network
- DSLE-5
- gymnasium
- Proximal Policy Optimization
AI-generated summary · Google Gemini · from 3 sources. How we write summaries →
