Autonomous + Human-Driven Mixed-Fleet Coordination

Regulatory Framework

The regulatory frame for mixed-fleet operation is layered across vehicle-classification, operational-oversight, functional-safety, and communications standards, and each layer presupposes architectural properties that the prevailing perception-and-motion-modeling pattern does not supply. SAE J3016 is the foundational classification standard; it defines automation levels L0 through L5 and explicitly contemplates that fleets will operate with members at different levels concurrently. The standard's text presumes that mixed-level operation is the steady-state condition for the foreseeable deployment horizon, not a transitional phase. NHTSA's Automated Vehicle Test Initiative (AV TEST) and the Standing General Order on Crash Reporting (SGO) impose disclosure obligations on operators of L2 and above systems; the SGO in particular requires that crash-relevant intent and behavior be reconstructable from the operator's records, which presupposes a lineage substrate the operator can produce on demand.

FMCSA Part 395 governs Hours of Service for human commercial drivers and mandates Electronic Logging Device compliance; in mixed autonomous-and-human-driven fleets the human-piloted unit is subject to HOS while the autonomous unit is not, and the fleet operator must coordinate the two populations under the same operational envelope. ISO 26262 governs functional safety for road vehicles; ISO 21448 (SOTIF, Safety of the Intended Functionality) governs the residual hazards arising from functional insufficiencies of the intended functionality, particularly under conditions where the system encounters scenarios the design did not fully anticipate — which is exactly the operational condition mixed-fleet coordination presents. V2X communications are governed by SAE J2735 (Dedicated Short Range Communications message set) and 3GPP C-V2X (cellular-vehicle-to-everything); ETSI ITS-G5 defines the European intelligent-transport-systems profile. Each of these communications layers presupposes a participant population in which Tier 1 cooperative broadcasts are exchanged with declared fidelity, but none specifies the architecture by which Tier 1 broadcasts are fused with Tier 2 partial signals and Tier 3 inferred attribution into a coherent admissibility evaluation.

Architectural Requirement

The architecture implied across these regulatory texts is a single intent-fusion substrate operating on observations of three structurally distinct fidelity classes and producing per-entity admissibility assessments that the planning stack consumes uniformly across the autonomous-and-human-driven population. Tier 1 is the cooperative broadcast: full-fidelity SAE J2735 BSM, J2735 SPaT, C-V2X CAM/DENM, ETSI ITS-G5 messages with verifiable attribution and declared semantics. Tier 2 is the partial-fidelity signal: turn signals, brake lights, hazard lights, horn, gesture, formation-position-relative-to-lead-vehicle, transponder-level partial broadcast. Tier 3 is the inferred attribution: trajectory analysis, gaze estimation where observable, formation pattern, historical-pattern-based attribution.

The substrate must produce a per-entity intent estimate with declared confidence and declared fidelity contribution by tier, support cross-tier corroboration that handles agreement and disagreement between tiers as first-class events, propagate lineage through the planning and actuation stack so that downstream decisions are auditable against the upstream observations that justified them, and operate uniformly regardless of whether the entity is an autonomous vehicle contributing on all three tiers or a human-driven vehicle contributing on Tier 2 and Tier 3 only. The substrate must also support graduated-cooperation transitions, in which a human-driven vehicle equipped with an aftermarket transponder upgrades from Tier 2/3 to Tier 1/2/3 contribution without the autonomous-vehicle population requiring re-architecting.

Why Procedural Compliance Fails

The conventional response to mixed-fleet regulation is procedural: an operator publishes a Voluntary Safety Self-Assessment, files SGO crash reports, maintains ISO 26262 functional-safety artifacts, and treats the human-driven population as a computer-vision target whose intent the perception stack infers from motion alone. The procedural model fails because the architectural pattern beneath it does not produce the structural properties the regulatory texts presuppose. A computer-vision target is not a coordination partner; the autonomous vehicle observes the human-driven vehicle's motion, infers intent from motion alone, and reacts accordingly. Tier 2 signals — turn signal, brake light, hazard, horn — are consumed informally if at all; the architecture has no first-class concept for them, no declared-fidelity slot for their contribution, no corroboration mechanism that combines them with motion-based inference.

The structural gap is most visible at the operational-domain boundary. The narrow-geography constraint that has held L4 deployment to specific urban districts and freight corridors largely reflects the difficulty of mixed-traffic operation under the computer-vision-target pattern. Geographies with high human-driver-population density — most of the United States outside specific neighborhoods, most of Europe outside dedicated AV-pilot zones — defeat the pattern at operational scale. SOTIF analysis of the resulting failure modes finds that the system encounters scenarios the design implicitly assumed away, namely scenarios in which the human-driven vehicle's intent is partially declared through Tier 2 signals the autonomous vehicle cannot first-class-consume. Procedural disclosure to NHTSA does not change this; it documents the failures, but the architecture continues to produce them.

What the AQ Primitive Provides

Three-tier intent fusion is the architectural primitive that supplies the structural properties the regulation presupposes. The composite admissibility evaluator runs continuously per neighboring entity. For each entity in observation range, the evaluator collects available Tier 1 broadcasts where present, Tier 2 signals as visual indicators, transponder data, and structured partial broadcasts, and Tier 3 inferences from motion-based attribution and historical pattern. Each tier carries declared fidelity. The output is a per-entity intent estimate with declared confidence and the per-tier contribution decomposition that downstream auditing requires.

The autonomous vehicle's planning consumes the estimate as a credentialed observation, weighting it appropriately for the planning horizon and the reversibility of the contemplated maneuver. Cross-entity coordination — formation maintenance, lane changes, intersection negotiation, merge-and-yield — operates through the same mechanism for all observed entities regardless of which tier they are contributing through. Cross-tier corroboration is first-class: when Tier 2 signals (turn signal active) corroborate Tier 3 inference (trajectory-implied lane change), the per-entity confidence rises and the planning stack is authorized to act on the higher-confidence estimate; when tiers disagree, the disagreement is recorded as a first-class event and the planning stack defaults to the more conservative interpretation. Lineage propagation flows through the entire stack so that NHTSA SGO reconstruction, FMCSA HOS coordination evidence, and ISO 26262/SOTIF post-event analysis can all draw from the same evidentiary substrate.

Compliance Mapping

The mapping from primitive to regulation is direct. SAE J3016 mixed-level operation is realized by treating each entity's automation level as a property of its tier-contribution profile rather than a population segregation; an L4 vehicle and an L0 human-driven vehicle differ in which tiers they populate and at what fidelity, not in whether they participate in the same admissibility substrate. NHTSA AV TEST and SGO disclosure are supported by the lineage propagation; crash-relevant intent reconstruction draws from the per-entity estimate's tier-decomposition history. FMCSA Part 395 ELD HOS coordination operates through the operator's fleet-management envelope, which consumes the same admissibility substrate to coordinate human-piloted and autonomous units under a unified operational policy. ISO 26262 functional-safety analysis treats the per-tier fidelity declarations as the basis for hazard analysis; ISO 21448 SOTIF treats cross-tier disagreement events as the substrate for residual-hazard characterization. SAE J2735 and 3GPP C-V2X Tier 1 broadcasts populate the Tier 1 slot directly; ETSI ITS-G5 messages do likewise in the European frame; Tier 2 partial signals and Tier 3 inferences populate their respective slots without architectural rewriting as the V2X penetration ratio shifts. The penetration ratio itself becomes a property of the operational design domain rather than a discrete capability boundary: the operator publishes the per-tier admissibility profile applicable to a given route or geofence, and the planning stack consumes the profile as a parameter rather than as an architectural switch. The result is that operational-design-domain expansion ceases to require re-clearing of the perception-and-motion stack and instead becomes a credentialed-policy-surface adjustment auditable against the operator's existing safety case.

Adoption Pathway

Operators building toward L4 commercial expansion can adopt the primitive at the planning-stack boundary without disturbing the perception-and-motion stack already cleared under existing safety cases. The pragmatic adoption sequence begins by wrapping the existing perception-and-motion output as the Tier 3 contribution into the composite evaluator, treating any cooperative broadcasts (V2X-equipped vehicles in the operational domain) as Tier 1, and treating turn-signal/brake-light/hazard detections as Tier 2. The wrapped system produces lineage evidence that materially exceeds the current procedural model, even before cross-tier corroboration logic is fully activated. As the operational domain expands, the operator activates cross-tier corroboration and graduated-cooperation transitions; aftermarket transponders deployed across human-driven fleets — gradually upgrading Tier 2 partial-fidelity signals toward Tier 1 cooperative broadcasts — extend the operational envelope without requiring full L4 capability across the human-driven population.

The compliance-driven adoption pattern is the same pattern V2X penetration has followed: regulators converge on architectures that map to the primitive's structure, and early adopters absorb the resulting domain-expansion advantage. Mixed-fleet operation is the steady state for the foreseeable deployment horizon; SAE J3016's text presumes it, NHTSA's reporting frame presumes it, FMCSA's HOS coordination presumes it, and the V2X penetration curve under SAE J2735 and 3GPP C-V2X presumes it. The narrow-geography constraint that has held L4 deployment to specific urban districts and freight corridors is not a perception problem and not a planning problem in isolation; it is an architectural problem at the layer where autonomous and human-driven units must coordinate as variants of the same primitive rather than as separate populations. Three-tier intent fusion is the primitive at that layer, and its adoption is positioned to compound as V2X penetration grows: each new V2X-equipped human-driven vehicle promotes from Tier 2/3 to Tier 1/2/3 contribution without architectural rewriting, and each new aftermarket transponder rollout extends the operational envelope incrementally rather than requiring a discrete capability step. Operators absorb the resulting domain-expansion advantage in the form of broader operational design domains, lower per-mile disengagement rates in mixed traffic, and SGO and ISO 26262/SOTIF evidentiary substrates that are structurally compatible with the regulatory direction of travel.