Edge AI compute

System-on-module compute with GPU or NPU acceleration for running perception and policy models onboard the robot.