Reproduce “Satisfied, Not Solving: Wireheading of Bounded Homeostatic Rewards in Reinforcement Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Chase J. Metoyer
The authors test whether a reset action that relieves internal needs disrupts learning in two MiniGrid environments. The action changes the needs read by a bounded intrinsic reward, not its formula. Agents with the action solved DoorKey-8x8 in 2 of 80 seeds and KeyCorridorS3R1 in 0 of 20, versus 71 of 80 and 20 of 20 without it. Return checks cover sampled, fully observed layouts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.