The most significant risk right now is training of the collected dataset. As the collected dataset will mostly be used to facilitate the accuracy in certain locations, where the collected data will actually improve detection accuracy will be significant to the meeting of the overall requirement. We manages those risks by trying to change the existing YOLO algorithm by slimming the output layers only for a limited amount of obstacles that might appear in hallway, and then, using public dataset will have a better effect.
We have not changed the block diagram as of now. Depend on the output accuracy of the training data, we might have to lower the accuracy and detection range due to hardware/software limitations.
The schedule is unchanged. We plan to deliver the model by Monday.