Robotic platforms are constrained by a feedback loop that should execute in typically less than 1/10 second on a low electric power budget. 

We have been developing custom algorithms using CUDA. We also off-loaded some computing load from the Mini-ITX board to a remote PC. Connecting the robot to a remote compute appliance helps in off-boarding compute-intense operations. More electrical power is available to operate the robot payload such as outdoor drive base, laser range finder, 4-DOF arm, etc.

Compute-intense computations are executed remotely. The API dynamically routes a stream either to the local machine or to a remote machines. The API also encapsulates compute nodes within XML web services.

The compute appliance option allows designers and researchers to work on compute intense algorithms such as visual SLAM.