Original title: Planning as Dynamics Relaxation: Hippocampal Recurrent Network Realizes Optimal Goal-Directed Navigation
Yuhang He, Junfeng Zuo, Tianhao Chu, Si Wu · August 2026
Scientists know the brain holds maps of physical spaces, but they do not fully understand how brain circuits plan paths around obstacles to reach a goal. Researchers designed a model of a hippocampal brain network to solve this puzzle. In this network, connections between neurons represent the likelihood of moving between different physical spots. If a wall or blocked corridor is in the way, the connection weight drops to zero. As an animal explores, the brain learns these patterns using behavioral-timescale synaptic plasticity, a known rule for adjusting brain connections. Once the network receives a target goal, its activity naturally settles, or relaxes, into a steady pattern. The local gradient of this activity points out the exact direction to walk. The authors found that this network navigates complex layouts efficiently and robustly. If the environment changes locally, the network only needs low-rank updates to adjust.
You are in a large, unfamiliar building and want to find the exit. Normally you would need to memorize the layout, but the building is too complex to remember every turn. Here, the hippocampal network is like the building: neuron connections represent possible paths, and walls are missing connections. The network's activity relaxes toward the exit neuron, automatically guiding you along the best route.
Understanding how the brain plans paths around obstacles reveals an efficient neural strategy. The finding that only small connection updates are needed when environments change points to adaptive learning. This principle could inspire navigation algorithms in AI and deepen our grasp of rational planning in the brain.
The work is limited to mathematical analysis and computer simulations; it does not include neural recordings or behavioral experiments. The model assumes a specific form of synaptic plasticity (BTSP) and a linearly-solvable MDP, which may not capture all complexities of real environments. Direct evidence that brains implement this exact computation is not provided.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Planning as Dynamics Relaxation: Hippocampal Recurrent Network Realizes Optimal Goal-Directed Navigation — Yuhang He, Junfeng Zuo, Tianhao Chu et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.