Reproduce “Planning as Dynamics Relaxation: Hippocampal Recurrent Network Realizes Optimal Goal-Directed Navigation” 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
Yuhang He, Junfeng Zuo, Tianhao Chu, Si Wu
The authors propose a hippocampal-style recurrent network that learns local transitions during goal-free exploration and uses the same connections for goal-directed planning. Recalling a goal clamps its location-coded unit, and activity settles into a field that a local readout converts to direction. In a simulated bug-trap maze, routes were 1.16× shortest legal length, despite incomplete connectivity; the result is model-based.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.