Path planning in the hippocampo-prefrontal cortex pathway: an adaptive model based receding horizon planner.

Four characteristic properties of human path planning strategy are course and fine planning, supervised planning, adaptation and robustness, and complexity reduction. These four characteristics are also observed in "model predictive controller" and its modified version, "receding horizon planner". We hypothesize that the human brain performs path planning tasks, literally like a receding horizon planner. The similarities between human brain and a receding horizon planner are: (1) hippocampus contains the course model and the parietal cortex is responsible for the fine model. (2) Replanning and trajectory tuning using the visual data in parietal cortex and prefrontal cortex is exploited in an adaptive restricted receding horizon. Prefrontal cortex plays the role of the supervisor. (3) Adjusting the sampling time of the planner is implemented based on changes in the complexity of the environment and tasks. This is in fact, the adaptation, which exists both in human behavior and in receding horizon planner. (4) The brain simplifies path-finding problems to reduce computational loads, exactly similar to what engineering controllers intend to do. The visual data is smoothed by clustering of obstacles, before performing any computational task. Finally, we have discussed the consequence of our hypothesis in Alzheimer disease as an optimal planning disorder. Based on some experimental data, Alzheimer patients have a reduced predictive horizon, making the system less robust and exposed to hazardous conditions in sophisticated environments. Patients with mild Alzheimer disease have little trouble with simple optimization problems; working memory of the prefrontal cortex is sufficient for this purpose. However, in complicated tasks, the brain needs huge extended memory. This memory is available through hippocampo-prefrontal pathway, which is to some extent disturbed in Alzheimer patients. We suggest that this fact may be a basis for future experimental diagnosis tests. We predict that Alzheimer patients should have problems with planning for far future; because they have a weak memory, insufficient for heavy optimization tasks, such as moving through moving obstacles in a dynamic environment. Alzheimer disease could be early detected by designing new tests in which the ability of patients to predict the future events is checked. These tests could be accompanied with a multi-step optimization problem. We believe that paying attention to this opinion may provide a good help in diagnosing Alzheimer disease in earlier stages. Surely, experimental studies are needed to validate our hypothesis.