Layering decisions for statistical techniques in autonomous system architectures

Determine where within layered autonomy architectures (behavior planning, path planning, motion planning, and feedback control) statistical techniques should be applied, and whether to use identical statistical and perceptual abstractions across layers or tailor layer-specific abstractions.

Background

Autonomous systems are typically structured in layers. The paper introduces statistical abstractions and conformal prediction-based verification and control, but does not prescribe architectural placement.

The authors explicitly state uncertainty about where and how to apply statistical techniques and whether abstractions should be standardized or customized per layer.