SpineIQ: Specification and Evaluation of a Deterministic Decision-Authority Kernel for Autonomous-Driving Runtime Assurance

Engineering notes

Key topics: autonomous driving, adas, control. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

SpineIQ: Specification and Evaluation of a DeterministicDecision-Authority Kernel for Autonomous-DrivingRuntime Assurance This repository contains the Release Candidate 1 (RC1) preprint and accompanying software reproducibility artifact for SpineIQ, a deterministic decision-authority kernel designed for autonomous-driving runtime assurance. Abstract: Autonomous-driving software distributes authority across planners, capability managers, runtime monitors, supervisory controllers, and fallback mechanisms. Although these components collectively determine whether vehicle behavior proceeds, authorization itself is rarely specified as an independent software responsibility. This paper investigates whether decision authority can be isolated as a bounded software artifact with implementation-agnostic semantics by design. We present SpineIQ, a deterministic decision-authority kernel that consumes contract-valid software state and produces exactly one of two observable outcomes: authorization of one abstract behavior or explicit refusal. The principal contribution is a Semantic Contract that defines authorization, refusal, capability eligibility, identity and replay semantics, timing and freshness, evidence-gated recovery, deterministic arbitration of simultaneous failures, and atomic state commitment. A frozen reference implementation is evaluated using deterministic scenarios, property-focused regression tests, differential verification against a separately implemented oracle, and an illustrative seven-cycle ADAS-shaped execution stream. The evaluated implementation completed 155 active scenario executions, 64 compatibility executions, and 50 pytest functions. Differential verification generated 40,000 candidates, adjudicated 35,260 of them, and observed zero disagreements between the reference implementation and the oracle within the evaluated domain. Artifact and Reproducibility: The attached archive (SpineIQ_Minimal_Driving_Kernel_V2_3_3_Contract_Reconciled.zip) contains the RC1 implementation, operational Semantic Contract summary, deterministic scenarios, compatibility suite, differential verifier, oracle, and environment metadata. A single script (reproduce.py) is provided to verify manifests and re-execute the release gate, test suite, and differential fuzzers to authorize reproducibility of the claims made in the manuscript.

Links and sources

Comments