So it's a very rough proof-of-concept right now. I mainly wanted to see if this type of robot could be done on an embedded board (the Jetson TK1). It works just barely, with the CV processes loading the CPU at almost 2x.
The heaviest tasks are the apriltag detector (findContours in opencv) and obstacle/floor pointcloud segmentation. Ideally this could all be done on the GPU, but the implementation is a bit non-trivial.
So it's a very rough proof-of-concept right now. I mainly wanted to see if this type of robot could be done on an embedded board (the Jetson TK1). It works just barely, with the CV processes loading the CPU at almost 2x.
The heaviest tasks are the apriltag detector (findContours in opencv) and obstacle/floor pointcloud segmentation. Ideally this could all be done on the GPU, but the implementation is a bit non-trivial.