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.

References

It is unclear in which layers we should be using statistical techniques, and if we should use the same statistical and perceptual abstractions across layers or if individual abstractions should be tailored to specific layers.

— Formal Verification and Control with Conformal Prediction  (2409.00536 - Lindemann et al., 2024) in Section 7, Open Problems and Future Directions

The key research question is where the assurance boundary should sit. If the shield operates only after trajectory generation, it may prevent collisions but not improve upstream reasoning. If it is differentiable and included during training, it can shape the planner but may be harder to certify.

— Planning-Oriented End-to-End Autonomous Driving: Architectures, Evaluation, and Emerging Paradigms  (2608.20111 - Guan et al., 20 Aug 2026) in Section 6.3, Runtime Assurance and Safety Envelopes