Mother-Child Vehicle (子母车) Research Review + Otsuka Solution: marsupial heterogeneous robot cooperation — cases, papers, results (working draft)

⚠️ A working draft under continuous verification. All paper and case links here are curl-tested reachable (HTTP 200). Two honesty notes: (1) arXiv 2202.08620 (Kirin) was withdrawn by its authors and is NOT SJTU's Baby Elephant — the sibling Fusion article wrongly cited it as Baby Elephant; this batch corrects it to the CJME review; (2) publisher pages (IEEE / ScienceDirect / ACM DOI) often return curl 403/202 from bot protection, not dead links, so this article cites arXiv abs pages / open-access PDFs preferentially. Industrial cases were WebFetch-checked for "is it really a carrier+child cooperation," with "solo quadruped inspection" explicitly excluded.

🌐 What this answers

The boss wants two things: (1) gather the existing cases, concept papers, and research on "子母车"; (2) combine with Otsuka's real need to produce a direction & solution. This article follows that order — research review first, then the plan.

It complements the siblings: the Fusion article covers the product-form taxonomy of "quadruped × AGV, three fusion forms"; Haulage selection covers "choosing a robot dog by payload tier"; this one focuses on the 子母车 line — its academic research + deployments + an Otsuka proposal.

🦘 First, the name: 子母车 = marsupial robotics

What Chinese industry calls "mother-child vehicle / car-dog cooperation" has a precise academic term — marsupial robotics: a larger carrier/mother robot carries a smaller passenger/daughter robot and releases it when reaching terrain the mother can't handle, by analogy to a kangaroo carrying its young in a pouch. This is exactly the architecture Otsuka needs: the carrier (AMR) hauls on flat ground, the child (robot dog) climbs the stairs. Searching this term surfaces a whole mature research area.

📚 1. Concepts & surveys (research foundation)

🔬 2. The most on-topic research

① Quadruped-carrier + child deployment criteria (the legged flagship of marsupial)

Marsupial Walking-and-Flying Robotic Deployment (NTNU + ETH, DARPA SubT team CERBERUS): a quadruped acts as the carrier carrying a drone, and when a shaft/steep terrain stops the ground system, an onboard algorithm decides "where and when to release" the child for focused exploration¹. Its 2025 follow-up adds task-driven map compression for ground-aerial cooperation under bandwidth limits².

Transfers to Otsuka: the carrier's "release the child only when terrain becomes impassable" deployment criterion is the algorithmic prototype of "release the dog only at the staircase."

② Deadlock-free scheduling of attachable heterogeneous carrier-shuttle AGVs (the industrial 子母车!)

Petri Net Modeling and Deadlock-Free Scheduling of Attachable Heterogeneous AGV Systems: splits heterogeneous AGVs into carrier (mother) and shuttle (child) that can flexibly attach / detach to haul together, modeling mother-child synchronization with Petri nets and guaranteeing deadlock-free scheduling⁵.

Transfers to Otsuka: nearly the exact scheduling model for Otsuka — how "carrier ferries child → decouple on arrival → child works solo → re-dock" avoids interlock and collision. The theoretical backbone for the "software orchestration" plan below.

③ Heavy legged-suspension carrier LEVA (the newest carrier-side hardware)

LEVA: A high-mobility logistic vehicle with legged suspension (ETH, 18-author team): a legged-wheel logistics vehicle with 85 kg payload, cost-of-transport as low as 0.15, climbs stairs, and autonomously picks up and loads cargo boxes⁴.

Transfers: it represents the trend of "the carrier itself crosses obstacles + self-loads" — if the carrier can climb stairs too, the mother-child division of labor gets redrawn (see Fusion article Form 1).

④ Dynamic docking on complex terrain (how the child "returns home")

Autonomous UAV-Quadruped Docking via Active Posture Alignment: a quadruped uses RL to actively stabilize its torso, "creating a level platform" on stairs/steep slopes for the child to land and dock autonomously³.

Transfers: the child's docking back to the carrier is the hardest part of 子母车; this "active posture alignment" method informs "how the dog climbs precisely back onto the AMR."

⑤ Tethered marsupial: the carrier powers the child continuously (beats endurance)

A line of work has the carrier power the child via a tether, extending the child's endurance from tens of minutes dramatically, and solves tether path/trajectory planning and physics simulation⁶⁷⁸.

Transfers: robot dogs typically run only 1.5–2 h; for a fixed short-radius cross-floor route, "carrier power / quick battery swap" can ease the child's endurance pain.

⑥ Heavy quadruped background (with a withdrawal warning)

Heavy quadrupeds research the ceiling of "how much the child can carry." SJTU's Baby Elephant (130 kg self-weight, 50–100 kg payload, motor-throttled hydraulics) is in the CJME review¹¹.

⚠️ Warning: the often-cited arXiv 2202.08620 is Kirin (50 kg electric quadruped) and was withdrawn by its authors — it is not Baby Elephant and its results are unreliable; don't use it for heavy quadrupeds (this batch fixes the sibling article).

🏭 3. Industrial deployments (real vs easily misused)

✅ The only verifiable "real car-dog cooperation" deployment: UISEE × State Grid Hangzhou (2025)

China's first "car-dog integrated": in the Binjiang IoT block of Hangzhou, the carrier = a UISEE UiBox L4 autonomous vehicle (10 km-class cross-zone transport on city roads), the child = a robot dog — the vehicle ferries the dog to a site, the dog dismounts to climb stairs and enter underground utility tunnels for last-100-m inspection (multispectral + gas + acoustic). Inspection coverage jumps from single points to 15 km², ~4× efficiency¹²¹³.

This is the only publicly verifiable "autonomous vehicle carrying a robot dog" deployment worldwide — the strongest case for telling Otsuka "someone has made this work." It's an inspection scenario with an outdoor vehicle carrier, but its architecture is structurally identical to Otsuka's "AMR carries dog across floors."

⚠️ For contrast: these are "solo quadruped + charging dock," NOT car-dog (don't misuse)

WebFetch-verified one by one — the following are often mistaken for "mother-child cooperation" but have no vehicle carrier; the dog walks and returns to a fixed charging dock on its own:

Lesson: a vendor having "a quadruped product + an auto-charging dock / a wheeled-legged machine" is easily spun as "car-dog cooperation." A deployment counts only if a carrier really ferries a child; otherwise it's ordinary quadruped inspection.

🗄️ The broad "子母车" ASRS (mature industrial goods, not quadruped)

Dense-storage four-way shuttle-carrier: a mother RGV runs the main rail to feed the child into the aisle, the child reaches deep into racks to store/retrieve pallets — Dematic Multishuttle[source], Inform/Yinfei[source] (A-share 603066, with an explicit "shuttle-mother + two-way shuttle-child" system). ⚠️ Wheeled, doesn't climb real stairs — shares only the "mother-child" label with a "quadruped child."

🎯 4. Mapping to Otsuka's need: direction & solution

Need recap (see Haulage selection): Otsuka currently hauls with flat-ground AGV/AMR that can't climb stairs/ramps; the boss wants robot dogs to fill the gap as a partnership pitch; per-item cargo weight is unknown and routes may be cross-floor.

Three transferable insights from the research:

1. The architecture is validated: marsupial / carrier-shuttle is a mature research area, and UISEE × State Grid has already made "car-dog integrated" work — the plan isn't speculation.
2. The hard part is orchestration, not hardware: mother-child attach/detach deadlock-free scheduling (arXiv 2508.00724), complex-terrain docking (arXiv 2509.21571), and endurance/power (tethered marsupial) are the real engineering problems — and these are exactly where software differentiates.
3. The carrier/child capability boundary is shifting: wheeled-legged carriers like LEVA self-cross obstacles, so "how far the carrier goes vs how much the child relays" must be re-cut per site.

Solution: dual-track in parallel, narrowed by real cargo weight

Track B engineering anchors (straight from the research above):

• Docking: AprilTag-guided self-climb / mechanical rail / a lift-ramp on the AMR, informed by "active posture alignment," to reach ≥99.9% on-site.
• Dispatch (RCS): use the carrier-shuttle Petri-net approach to choreograph "AMR drive → arrive → wake dog → dog climbs → dog returns → dock → AMR departs" into a deadlock-free automated workflow, aligned with VDA5050 to plug into Otsuka's existing AGV master — the core differentiation of in-house software (buy the hardware; the orchestration and cooperation are ours).
• Endurance: for fixed short-radius cross-floor routes, prioritize "carrier quick-swap/power station" to ease the dog's 1.5–2 h endurance.

Phased PoC roadmap (recommended)

1. Phase 0 · ask 5 numbers: per-item cargo weight/size, cross-floor or not, step count & slope, takt, existing AGV model — decides Track A vs B.
2. Phase 1 · Track A minimal validation: one B2-W (via Japan distributor TechShare) runs "stair haulage + in-house dispatch software"; low failure cost, prove the loop first.
3. Phase 2 · Track B mother-child docking: add "AMR carrier + docking + RCS orchestration" on top of A's software stack, benchmarking the UISEE × State Grid "car-dog integrated" for an in-plant version.
4. Phase 3 · VDA5050 networking: plug the whole workflow into Otsuka's existing AGV dispatch, forming a "software × hardware" combined offering to pitch.

Honest risks (state plainly to the client)

• Only one real car-dog deployment exists worldwide (and it's outdoor inspection, not in-plant haulage) — this is frontier, not a shelf-mature product; advance via PoC, don't over-promise.
• A single robot dog still can't carry heavy goods (>40 kg) up stairs — heavy loads need "carrier on flats + dog relays stairs" or split trips.
• Docking and cross-floor offline autonomy are engineering hard points — field-test docking success rate and lost-link return.

⚠️ To verify / to add

1. Real cargo weight + route (most critical): decides Track A vs B and narrows the plan to specific models and cost.
2. Want me to dig deeper into the UISEE × State Grid "car-dog integrated" case (technical detail, in-plant transferability, contacting the integrator)?
3. Publisher full texts: due to bot-blocks, this article cites only arXiv/open PDFs; if you have institutional access and want formal DOIs, I'll compile a DOI list.
4. Want this research + plan as a one-page client proposal (docx/ppt)? I can add a "mother-child vehicle direction" chapter onto Section 12 of the business-plan template.