How does DEPLOY classify autonomy boundaries across physical AI?

The four-way autonomy-boundary taxonomy

Autonomy boundaries across physical AI sort structurally into four distinct tiers. The taxonomy first surfaced editorially in the surgical cluster as the 5th architectural axis (autonomy-boundary added to FDA-clearance-posture + clinical-domain + commercial-model + orthopedic sub-cohort triangle); it now operates as cross-cohort architectural classification within DEPLOY's framework discipline. The four tiers:

1. Autonomous-execution: the machine executes the operation under AI control with operator supervision. The operator approves a plan and supervises execution; the machine performs the operation. Monogram mBos (KUKA-based robotic arm executes the bone cut under AI control within surgeon-approved CT plan and supervision) is the canonical worked example. The robot cuts. The surgeon supervises.

2. AI-augmented operator-controlled: the operator executes the operation; software plans + tracks + bounds. The AI provides patient-specific planning, intra-op tracking, and motion bounds within the planned envelope; the operator physically executes the operation. Stryker Mako + Smith+Nephew CORI + Zimmer Biomet ROSA (orthopedic sub-cohort triangle) anchor this tier. The surgeon cuts. The software bounds.

3. Replacement-robotics teleoperated: the operator teleoperates remote actuators via console; the robotic platform replaces the operator's direct manipulation of instruments. Intuitive da Vinci + Medtronic Hugo + J&J Ottava + CMR Versius anchor this tier in soft-tissue surgical robotics. The surgeon teleoperates. The robotic arms execute the teleoperation in real-time.

4. Assistive co-pilot: the operator directly manipulates instruments with augmentation from the system. The system doesn't replace direct manipulation; it augments the operator's existing technique. Moon Surgical Maestro (Laparoscopy Co-Pilot) anchors this tier. The surgeon directly manipulates laparoscopic instruments. The co-pilot augments.

Per how-deploy-verifies methodology editorial, the four-way distinction is editorial signal at the verification-posture layer: trade-press coverage flattening physical AI systems into uniform "autonomous" or "AI-powered" framing collapses structurally distinct positioning that institutional partners evaluate at primary-source-verified depth.

---

Cross-cohort application: surgical

The surgical cluster operates the canonical worked-example application of the four-way taxonomy. All four tiers surface across verified-entity anchors per the surgical cluster framework:

• Autonomous-execution (surgical): Monogram mBos. Semi-autonomous version FDA-cleared March 17 2025. KUKA-based robotic arm executes bone cut under AI control within surgeon-approved CT plan and supervision. Pre-commercial maturity (no units sold). Critical cap-flag: "semi-autonomous" is Zimmer Biomet + trade-press-sourced framing, NOT FDA-letter-verbatim; Monogram's own PR uses "robotic-assisted TKA."

• AI-augmented operator-controlled (surgical): Stryker Mako (large-footprint CT-based robotic-arm-assisted; multi-procedure cleared knee + hip + spine via K241517 + shoulder via K242373); Smith+Nephew CORI (handheld imageless; total + partial + revision knee; CRITICAL HONESTY DISTINCTION hip is NAVIGATION-ONLY NOT robotic); Zimmer Biomet ROSA (mid-size cross-domain knee + hip + brain SEEG/DBS + shoulder; CRITICAL CORRECTION ROSA Spine LEFT ZB in 2022 ZimVie spinoff; Monogram autonomy enters via SEPARATE acquisition NOT ROSA).

• Replacement-robotics teleoperated (surgical): Intuitive da Vinci (market-leader installed-base 11,395 da Vinci + 1,041 Ion systems per SEC 10-Q); Medtronic Hugo (large-prime IDE-pivotal pathway; commercial in CE-Mark regions); J&J Ottava (De Novo submission January 2026; FDA "not authorized to be marketed"); CMR Versius (UK-headquartered modular soft-tissue surgical; international commercial EU + UK + Australia + India; US FDA state cap-flagged).

• Assistive co-pilot (surgical): Moon Surgical Maestro (FDA 510(k) K240598 cleared June 2024; smaller-footprint OR integration; commercial-niche positioning).

Verification criteria per tier within surgical cohort: who physically executes the operation + who supervises + what FDA clearance scope covers + what cap-flag attaches to autonomy-classification claims. The tier classification matters editorially because the verification posture differs structurally: "the robot cuts" + "the surgeon cuts with software bounds" + "the surgeon teleoperates the robotic arms" + "the surgeon directly manipulates with augmentation" are not the same claim.

---

Cross-cohort application: drone

The drone cohort surfaces the four-way taxonomy with cohort-specific cap-flag depth. Per the autonomous drones cluster framework, the new-defense AI-first / legacy-prime / commercial-civilian three-class cohort intersects with the autonomy-boundary classification:

• Autonomous-execution (drone): new-defense AI-first cohort entries operate autonomous mission execution under operator-approved mission plans. Anduril Ghost + Helsing HX-2 + Shield AI V-BAT anchor the autonomous-mission-profile tier per cluster framework: AI-first autonomy stacks built as core product; operator approves mission plan; AI executes mission profile (target loiter; route execution; mission completion). Per autonomous drones cluster, specific autonomy-execution claims at scaled-mission-execution depth operate at lower verification posture pending primary-source confirmation per Pentagon-opacity-grade rigor.

• AI-augmented operator-controlled (drone): commercial-civilian cohort + some new-defense entries operate AI-augmented control where the operator directs the flight envelope and AI augments specific functions (obstacle avoidance + autonomy stack assistance + return-to-launch). Skydio Autonomy Engine (deployed across commercial-civilian drone platforms) + Wing OpenSky + Zipline Autonomy anchor commercial-civilian AI-augmented operator-controlled tier per registry brain-substrate framework.

• Replacement-robotics teleoperated (drone): legacy-prime cohort entries operate primarily as remotely-piloted platforms; the operator pilots the platform continuously via console; AI augments specific functions but does not execute autonomous mission profiles. General Atomics MQ-9 Reaper + Baykar Bayraktar TB2 anchor the legacy-prime remotely-piloted tier. Per cluster framework: MQ-9 + TB2 are recorded as remotely-piloted, NOT autonomous, regardless of marketing framing. Per cap-flag-as-trust-signal, the framework resists "drone equals AI" inflation. Attack-FPV sub-category (Neros Archer) also operates at remotely-piloted tier.

• Assistive co-pilot (drone): not a primary tier in the drone cohort (drones don't operate human-direct-manipulation primarily); the closest analog is commercial-civilian platforms with operator-mode + autonomy-mode switching.

The drone cohort intersection with the four-way taxonomy surfaces the remotely-piloted-vs-autonomous honesty discipline editorially: legacy-prime AI-augmented platforms get framed as autonomous in trade-press; the framework resists the inflation per primary-source-anchored verification.

---

Cross-cohort application: maritime

The maritime cohort applies the four-way taxonomy at the new-defense-vs-legacy-prime + USV-vs-AUV intersection per maritime cluster framework:

• Autonomous-execution (maritime): new-defense AI-first subsea production tier. Anduril Ghost Shark (A$1.7B RAN production contract September 2025; sub-killer ASW mission profile; Lattice-wired) anchors autonomous-mission-profile execution at the production tier; Anduril Dive-LD (USN UUVRON-1 fielded; 6,000m operating depth; Lattice-wired) at the fielded tier. The new-defense AI-first cohort extends autonomous-mission execution from aerial into subsea per cluster framework.

• AI-augmented operator-controlled (maritime): surface USV cohort entries operate operator-supervised execution with AI augmentation for navigation + mission planning + autonomous transit. Saronic Corsair (new-defense AI-first surface USV; $1.75B Series D at $9.25B March 2026; Corsair $392M ceiling with ~$197M obligated vessel-unnamed) + Saildrone USV (captive data-as-a-service ocean-research-focused) anchor surface AI-augmented operator-supervised positioning.

• Replacement-robotics teleoperated / programmed-mission (maritime): legacy-prime subsea family entries operate at the teleoperated or pre-programmed-mission tier. HII REMUS family (300 + 600 + 6000; Yellow Moray USS Delaware 2025 verified fielded anchor) anchors legacy-prime subsea teleoperated/programmed-mission positioning. The framework cap-flags HII REMUS at the operator-supervised execution tier rather than autonomous-mission-execution tier per the new-defense subsea distinction.

• Vehicle-not-conglomerate gradient (maritime): Kongsberg HUGIN AUV + Exail DriX USV operate vehicle-scope-not-conglomerate entity scoping with autonomy-classification at the AI-augmented operator-controlled tier; specific autonomy execution depth cap-flagged pending maker primary-source confirmation.

The maritime four-way taxonomy intersection makes the new-defense-vs-legacy-prime architectural distinction operate at the autonomy-boundary layer in addition to the procurement-lineage layer: new-defense AI-first cohort operates more autonomous-mission-execution capable; legacy-prime cohort operates more teleoperated/programmed-mission positioned.

---

Cross-cohort application: humanoid

The humanoid cohort applies the four-way taxonomy at the consumer-vs-industrial sub-cohort + Chinese-vs-Western geographic axis per humanoid cluster framework + consumer-vs-industrial humanoid sub-cohort architecture:

• Autonomous-execution (humanoid): aspirational tier across the cohort; no humanoid platform operates at verified autonomous-execution maturity at consumer or industrial scaled-throughput tier as of mid-2026. Per consumer-vs-industrial sub-cohort architecture, consumer humanoid commercial maturity at 2027-2028 aspirational; industrial humanoid at limited-pilot + selective scaled-throughput verified anchors. Per what-is-physical-intelligence brain-provider entity anchor, foundation-model-for-robotics integration enables autonomous-execution capability at the brain-provider integration layer; integration-to-deployment translation operates at lower verification posture.

• AI-augmented operator-controlled (humanoid): industrial humanoid pilot deployments operate at AI-augmented operator-controlled tier with explicit teleoperation disclosure. Figure 03 at Catalyst Brands Reno verified commercial customer + Apptronik Apollo Mercedes-Benz + GXO + Atlas Hyundai Metaplant + Agility Digit GXO Flowery Branch 100,000-tote scaled-throughput verified anchor (the canonical scaled-throughput verified deployment in the humanoid cluster).

• Replacement-robotics teleoperated (humanoid): consumer humanoid demonstrations + most public-demo footage operates at the teleoperated tier explicitly. 1X NEO's Expert Mode teleop framing sets the disclosure-depth bar editorially: 1X explicitly discloses that consumer-shipped NEO demonstrations include human-teleoperated functions. Tesla Optimus public demonstrations operate at teleoperated tier per public reporting; specific Optimus autonomy-execution claims at scaled-deployment depth cap-flagged pending primary-source confirmation. Chinese humanoid sub-cohort public demonstrations operate at demonstration / teleoperation / choreography scope per consumer-vs-industrial humanoid sub-cohort architecture editorial framing.

• Assistive co-pilot (humanoid): not a primary tier in the humanoid cohort as currently structured (humanoid platforms don't operate operator-augmentation tier primarily); closest analog is research-tier teleoperation-with-AI-augmented-policy positioning at the brain-provider integration layer.

The humanoid four-way taxonomy intersection surfaces the editorial discipline at the autonomy-execution layer: 1X NEO's Expert Mode disclosure operates as canonical worked example of cohort-member operational disclosure depth; cohort members whose autonomy claims do not include comparable disclosure carry cap-flag tier on capability questions per cluster framework.

---

Cross-cohort application: sidewalk-delivery

The sidewalk-delivery cohort applies the four-way taxonomy with primary positioning at operator-supervised L4 + remote-teleoperator edge cases per sidewalk delivery cluster framework:

• Autonomous-execution (sidewalk-delivery): primary operational tier per cohort framework. Starship Technologies + Serve Robotics operate sidewalk-autonomous-execution at operational deployment maturity. Operator-supervised L4 autonomy: the robot navigates the sidewalk autonomously within the operational envelope (geofenced commercial campuses; specific routes; mapped sidewalk infrastructure).

• AI-augmented operator-controlled (sidewalk-delivery): not a primary tier; the operator-supervised L4 autonomy is the dominant operational mode.

• Replacement-robotics teleoperated (sidewalk-delivery): remote-teleoperator edge cases + intervention modes. Remote-teleoperators handle edge cases that exceed the L4 operational envelope (intersection navigation in unmapped contexts; obstacle resolution; recovery from navigation failures). The teleoperator intervention is structurally distinct from primary operational mode but is a load-bearing component of the deployment.

• Assistive co-pilot (sidewalk-delivery): not a primary tier.

The sidewalk-delivery cohort cap-flag discipline editorially: trade-press coverage that flattens sidewalk-delivery operations into uniform "fully autonomous sidewalk delivery" framing collapses the operator-supervised L4 + remote-teleoperator intervention distinction. Per cap-flag-as-trust-signal, the verification posture appropriate to sidewalk-delivery claims surfaces the operator-supervised + teleoperator-intervention reality honestly.

---

Verification criteria per tier

The four-way taxonomy operates with structurally distinct verification criteria per tier:

Autonomous-execution verification criteria: who approves the operation plan + what AI architecture executes + what operator-supervision posture covers + what failure-mode disclosure exists + what regulatory-clearance scope authorizes autonomous execution. Per surgical worked example: Monogram mBos FDA-cleared semi-autonomous version March 17 2025 with KUKA-based execution + surgeon-approved CT plan + surgeon supervision. The verification posture requires primary-source-anchored confirmation at the FDA letter granularity; trade-press "semi-autonomous" framing operates at lower verification depth.

AI-augmented operator-controlled verification criteria: what AI augmentation scope covers + what motion bounds the software enforces + what operator-execution depth remains + what FDA clearance covers the augmentation scope. Per orthopedic sub-cohort triangle worked example: Mako + CORI + ROSA software plans + tracks + bounds; surgeon makes the cuts. The verification posture requires confirming the operator-execution scope explicitly; trade-press "autonomous surgical robot" framing collapses the AI-augmented vs autonomous-execution distinction.

Replacement-robotics teleoperated verification criteria: what teleoperation latency + what operator-console depth + what direct-manipulation replacement scope + what verified-installed-base evidence exists. Per da Vinci worked example: 11,395 da Vinci + 1,041 Ion systems per SEC 10-Q; surgeon at console teleoperates robotic arms in real-time. The verification posture requires installed-base + procedure-volume primary-source attribution; trade-press "Intuitive's dominance" framing should be read against installed-base differential per cluster framework.

Assistive co-pilot verification criteria: what augmentation scope covers + what direct-manipulation operator-execution remains + what FDA-clearance authorizes the augmentation. Per Maestro worked example: FDA 510(k) K240598 cleared June 2024; surgeon directly manipulates laparoscopic instruments with co-pilot augmentation; smaller-footprint OR integration; assistive-vs-replacement positioning preserved.

---

Aggregator-drift patterns per tier

Each tier attracts structurally distinct aggregator-drift patterns:

Autonomous-execution drift patterns: trade-press inflation of operator-supervised execution as fully-autonomous execution; framing of pre-commercial autonomous-execution capability at commercial-deployment depth; collapse of regulatory clearance scope distinction (semi-autonomous vs fully-autonomous). Cap-flag discipline surfaces the verification depth honestly per autonomous-vs-aspirational-autonomous distinction.

AI-augmented operator-controlled drift patterns: trade-press inflation of AI-augmented systems as autonomous; collapse of operator-execution depth distinction; conflation with autonomous-execution tier. Per orthopedic sub-cohort triangle editorial framing applied consistently: the surgeon makes the cuts; the AI bounds the cuts; the AI does not make the cuts.

Replacement-robotics teleoperated drift patterns: trade-press collapse of teleoperated platforms with autonomous platforms; framing of teleoperated systems as AI-powered; conflation with AI-augmented or autonomous-execution tiers. Per drone cohort editorial framing: MQ-9 + TB2 are recorded as remotely-piloted, NOT autonomous, regardless of marketing framing.

Assistive co-pilot drift patterns: trade-press collapse of assistive-co-pilot positioning into replacement-robotics or AI-augmented framing; loss of direct-manipulation-with-augmentation distinction. Per Maestro editorial framing: surgeon directly manipulates laparoscopic instruments; co-pilot augments; the system does not replace surgeon manipulation.

---

Why the taxonomy matters

The four-way autonomy-boundary classification matters editorially because it operates as cross-cohort architectural distinction at the verification-posture layer where institutional partners audit. Insurance underwriting requires distinguishing autonomous-execution coverage from AI-augmented coverage from teleoperated coverage from assistive coverage at primary-source-verified depth. Data sharing permissions require distinguishing autonomous-execution data flows from AI-augmented operator-controlled data flows from teleoperated data flows from assistive co-pilot data flows. Cross-device interoperability requires identifying which autonomy tier each device operates at primary-source-confirmed level.

The framework operates recursively at the autonomy-boundary classification layer just like it operates at the verified-vs-claimed framework tier + the captive-vs-third-party brain-provider gradient + the form-factor-cell sub-cohort treatment layer. Cross-cohort framework reproducibility at every layer of the cohort architecture is the discipline.

For the canonical autonomy-boundary worked examples, see the surgical cluster editorial canon: Monogram mBos (autonomous-execution) + Stryker Mako + Smith+Nephew CORI + Zimmer Biomet ROSA (AI-augmented operator-controlled) + Intuitive da Vinci + Medtronic Hugo + J&J Ottava + CMR Versius (replacement-robotics teleoperated) + Moon Surgical Maestro (assistive co-pilot). For the methodology editorial canonical reference, see how DEPLOY verifies. For the cross-cohort cluster framework canonical references, see the surgical robotics cluster + the autonomous drones cluster + the maritime robotics cluster + the humanoid robots cluster + the sidewalk delivery cluster.