Early stage 1: build prototype
Open Source Evolutionary Robotic Quadruped

Inputs:
360 degree field of view for CV
Pressure sensors
Accelerometers and gyroscopes
Temperature sensors

Outputs:
5 degrees of freedom legs using brushless dc motors and o-drive controllers (Leg extension and retraction is degree of freedom)
Dynamic weight shifting in torso controlled using brushless dc motors and o-drive controllers

From deep learning inference classify scenario (ground gradient and properties, moveable objects on ground and their properties, moving objects and their properties).

Based upon this data have algorithms for optimal motor movements for getting to input coordinates.

Have scoring metric on how well the quadruped succeeded at this and then us GA to alter parameters slightly for next test. Update BKM algorithm for each classification object.

Frequently sync this data to a shared database where it will retrain the optimal network for inference and traversing.