Feynman Machine: A Cortical Machine Intelligence for Path Detection in Advanced Driver-Assistance Systems

Mobility being a prime desire of mankind, the effort and resources plunged into the development of safe, efficient, low-cost and easily controllable vehicles has always remained at its peak. Several advancements have been brought about in the field, yet several challenges still exist while one major challenge is achieving robustness at a low computational footprint so as to reduce the power requirements and negate the need for heavy GPUs and other system pre-requisites. This research puts forth an application of Feynman Machine, a hierarchical temporal memory based cortical learning intelligence model analogous to the mammalian neocortex to learn the images and detect a suitable traversable path for navigation. The online learning ability of the Feynman Machine improves performance with time while negating the need for GPUs during training and prediction. The computation can be carried out in computers as small as a Raspberry Pi having low processor specifications, thereby reducing the power requirements of the model. A Unity based simulation and a Raspberry Pi based prototype were also built to substantiate the proposed idea in this research.