English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

DSLE: A Learning Environment for Dark Souls Boss Encounters

Forum topic · 小凯 · 2026-08-11

Summary

The Dark Souls Learning Environment (DSLE) is a containerized platform that turns all 22 boss encounters from Dark Souls: Remastered into benchmarks for game-playing agents, exposed through a Gymnasium-style interface. Each environment step executes a real action against the running game, combining real-time combat, high-dimensional visual input, and sparse terminal rewards. For controlled comparisons, the authors define DSLE-5, a representative five-boss subset covering melee combat, spatially constrained arenas, environmental hazards, multi-target fights, and a fast final boss, recommended as a starting suite. On DSLE-5 they evaluated a random policy, an expert system, an evolutionary baseline, and PPO and DQN agents trained from visual input. Expert systems and evolutionary baselines each defeated the tutorial boss (Asylum Demon) with peak win rates of 63% and 43%, but none of the five methods defeated the other four bosses. PPO and DQN showed no measurable learning within budgets of tens of wall-clock hours per run (max 0.33% win rate on the tutorial boss, 0% elsewhere). A broader study running the evolutionary baseline across all 22 encounters, with all 50 levels of advantageous attributes, won only a few additional early-game bosses. Failure cases ranged from sub-10-second deaths to nearly minute-long stalemates, reported via survival time and damage dealt rather than win rate alone. arXiv: 2508.03798.

We introduce the Dark Souls Learning Environment (DSLE), a containerized platform that presents all 22 boss encounters of *Dark Souls: Remastered* as game-playing agent benchmarks through a Gymnasium-style interface. DSLE combines real-time combat, high-dimensional visual input, and sparse terminal rewards, with each environment step being a real action executed against the running game.

  • Paper: arXiv:2508.03798
  • Authors: Derin Gezgin, Jim O'Connor, Tanner Goodwin
  • Field: Machine Learning
  • Published: 2026-08-11
  • Key points

  • DSLE wraps the full game as a benchmark: 22 boss encounters, real-time combat, visual observations, and sparse terminal rewards via a Gymnasium-style API in a containerized setup.
  • DSLE-5, a representative five-boss subset, is proposed as the standard starting suite, spanning: a melee fight, a spatially constrained arena, an environmental-hazard fight, a multi-target fight, and a fast final-boss fight.
  • On DSLE-5, the authors evaluate a random policy, an expert system, an evolutionary baseline, and PPO and DQN agents trained from visual input.
  • Results: the expert system and evolutionary baseline each defeated the tutorial boss (Asylum Demon) with peak win rates of 63% and 43%, respectively. None of the five methods defeated the other four bosses in DSLE-5.
  • PPO and DQN agents showed no measurable learning within budgets that already required tens of wall-clock hours per run (at most 0.33% win rate on the tutorial boss, 0% elsewhere).
  • A broader study ran the evolutionary baseline on all 22 encounters; even with all 50 levels of advantageous attributes, it won only a few additional early-game bosses.
  • Failure cases ranged from deaths in under 10 seconds in narrow multi-target encounters to stalemates lasting nearly a minute with almost no damage dealt. Results are therefore reported using survival time and damage dealt, not just win rate.
DSLE highlights how far current RL methods are from mastering real-time 3D action games with sparse rewards, and offers a standardized, reproducible suite for future agent research.

--- *Auto-collected on 2026-08-12*

Tags

#machine-learning#reinforcement-learning#game-ai#benchmark#dark-souls#ppo#dqn#gymnasium

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/178633354