Updated

Marsupial-Shuttle Warehouse Transport on SEER's System — A Feasibility Study (downloads-center first-hand check × Japanese support × SRC heterogeneous compatibility × GO2/B2 fit · working draft)

仙工SEERM4SRC子母车仓储运输VDA5050日语宇树GO2B2大冢商会方案
On this page

⚠️ A working draft under continuous verification. All links here are curl-tested reachable (HTTP 200). Three caveats first: (1) SEER's official software/controller pages and downloads center are first-hand; D1/D2 quadrupeds, wheel-leg variants, WRC exhibits are media-sourced (zhineng518 / leaderobot) — credible but worded per media. (2) The specific third-party AGV brands/models M4 supports, and the dog-specific interface (whether joint bus is EtherCAT/CAN-FD), are NOT published — every concrete compatibility claim here is marked "verify with SEER / PoC." (3) The naming collision must be kept straight: Unitree D1 = a 6-DOF robotic ARM (not a dog); SEER's D1/D2 = quadruped dogs, and they are "partner-built hardware + SEER's brain," NOT SEER-manufactured units.

🌐 What this answers

A very concrete question: can SEER's system build "marsupial-shuttle warehouse transport" — a mother car carrying a child dog across zones, the dog dismounting to do the final stair-climb/squeeze-through haul, all under one dispatch? This article checks the boss's four named points against first-hand evidence, then gives the marsupial build and its gates.

Sister articles: SEER M4 landing-path deep-dive (interface × connecting Otsuka's AGV × procurement), quadruped × AMR/AGV integration cases (who actually wired up with whom), marsupial-shuttle research (academic + plan), self-built middleware architecture (pure-DIY fallback). This one does just the "marsupial shuttle on SEER's system" feasibility.

🧭 One-line overview

The path holds — and it's the most ready-made domestic option: SEER M4 as the unified dispatch brain (RCS), with the mother car (an AMR/AGV carrying SRC or VDA5050) and the child dog (a quadruped with an SRC-5000 brain) queued and tasked together in M4 — SEER has already put "quadrupeds running its control system" into M4 unified dispatch. But three hard gates remain: whether Otsuka's existing AGVs support VDA5050 (the biggest variable), SEER's lack of a local agent in Japan, and the still-self-built dock mechanism.

📥 First-hand check ①: SEER downloads center (verified downloadable)

The boss asked to "download the system/manuals for first-hand verification." Done:

  • seer-robotics.ai/zh/download curl 200 — a genuine "Downloads Center" ("abundant product info, video and software downloads"), with Product Catalogues / Product Videos categories1. Individual PDF filenames are JS-rendered (curl can't list them) and need a browser to download per-file, but the entrance and categories are confirmed.
  • An even better find for Otsuka: there is a dedicated Japanese downloads center seer-robotics.ai/jp/download (curl 200)2— the JP team can pull Japanese materials directly.
  • Companion first-hand pages: software overview3, SRC controllers4, M4 dispatch (en/fleet)5, Nebula selection platform7.

Suggested action: have the JP team enter /jp/download, pull the M4 catalogue and SRC controller manual, and focus on the concrete config for third-party protocol onboarding / VDA5050 / quadruped interface — the fastest way to turn "compatibility" from media wording into first-hand evidence.

🗾 First-hand check ②: Japanese support — a natural plus for Otsuka

  • Official multilingual, confirmed: seer-robotics.ai's language switcher lists 9 languages, incl. 日本語 / English / 한국어 / Deutsch / Italiano / Français / Svenska / Español / Português, plus a dedicated JP site seer-robotics.ai/jp3.
  • ⚠️ Soften this (honest): Japanese is confirmed at the site and corporate-materials level; but "the M4 product UI itself ships a Japanese locale" was NOT directly verified (the M4 micro-site only exposed CN/EN). → Tell the client "SEER officially offers a JP site and Japanese materials," not "the M4 UI is Japanese-localized" — the latter awaits SEER confirmation or an on-site check.
  • And one more honest point: a JP site ≠ local support in Japan. SEER's only overseas subsidiary is in Germany; there is no local agent in Japan (the JP site's inquiry still routes to HQ) — see the SEER M4 landing-path article. This actually leaves room for a "Japan local integrator": Otsuka buys SRC/M4 licenses and does the upper-layer integration and Japanese on-site support itself.

🔧 First-hand check ③: retrofit SRC to make heterogeneous dogs/AMRs/AGVs compatible

This is the key to whether a "marsupial shuttle" can be assembled from heterogeneous hardware — and SEER's signature logic: a unified "brain."

  • The concept holds: any third-party AGV/AMR/dog with a SEER SRC controller inside plugs seamlessly, cross-model into M4 unified dispatch; SRC handles 3D-SLAM mapping, mm-level localization and obstacle avoidance, and the upper layer auto-generates the M4-standard interface4.
  • Ecosystem scale (correcting the doc's "800"): the official figure is "over 1,000 customers"4, with third-party reporting at 1,300+ partners — the doc's "800" is likely confused with the SRC-800 model; use 1,000+ customers / 1,300+ partners.
  • Selection (correcting the doc's SRC-1100):
    • entry SRC-800 / 880 (twin-wheel differential light AMR, the high-volume price-puller);
    • general/forklift SRC-2000 (single-steer / omni / jacking / forklift);
    • SRC-1100 = a dual-steering-wheel controller (EtherCAT, high real-time) — NOT a quadruped/embodied controller (the doc treating it as a dog controller is wrong);
    • legged/embodied ONLY uses SRC-5000: officially "the world's first integrated embodied-intelligence controller," baking AGI capability into an industrial controller, for quadruped/humanoid — this is the one that serves as a dog's brain4.
  • Third-party dogs have actually done it: SEER "together with ecosystem partners" launched four-leggers running SEER's control systemD1 navigation model (3D-SLAM logistics delivery, IP54, 0–40 °C), D2 patrol model (thermal/pickup/strobe), wheel-leg D1-W/D2-W (CeMAT exhibit)89.
  • ⚠️ Errata (more precise than the doc): these D1/D2 are NOT SEER-built units — the official wording is "together with partners" / "running SEER's control system," i.e. partner builds the dog, SEER supplies the brain (SRC-5000 + navigation), the hardware maker unnamed, media-named D1/D2. And they are entirely unrelated to Unitree's D1 (a 6-DOF robotic ARM, 2.37 kg / 500 g payload, not a dog)12— keep the two "D1"s straight.

On cost: SEER listed on HKEX on 2026-06-24 (06106.HK); the prospectus discloses SRC controller average price fell from RMB 25,900 (2023) to RMB 10,700 (2025), −58.7%, mainly because the high-volume SRC-880 dragged the average down1314. The tiered prices (SRC-800/880 ≈ RMB 3k–8k, SRC-2000 ≈ RMB 12k–25k, high-end SRC-5000 etc. ≈ RMB 30k–60k+) are secondary-market estimates, NOT official quotes — get a direct quote. On top of hardware there is usually a software license fee (navigation algorithms / M4 node license).

🐕 First-hand check ④: Unitree GO2 / B2 fit with M4

Can the boss's two named Unitree dogs serve as the "child dog"? Checked item-by-item against the official site:

Dimension GO2 (light-load starter) B2 (heavy-load mainstay)
2nd-dev / prerequisite to reach M4 Only EDU (and X industry ver.) opens SDK; Air/Pro consumer locked11 Natively fully-open SDK ("solid-circle" tier)10
Whole-machine weight ~15 kg ~60 kg (B2-W ~85 kg)
Haulage/load rated ~7 kg; official "peak ~10 kg" is a peak, not haulage continuous walking load >40 kg; standing ≥120 kg is static, NOT haulage10
Obstacle/environment stairs/slope (unloaded nominal); no IP rating published 40 cm step, slope >45°, IP67, −20 to 55 °C10
Compute/IO EDU can mount Jetson Orin i5/i7 + up to 3× Orin NX, 4× gigabit ethernet + 4× USB3.010
Way to reach M4 EDU self-built middleware / bolt-on SRC-5000 Back-mounted power feeds an SRC-5000 most cleanly, or self-write a VDA5050 bridge
  • Fit conclusion: neither can reach M4 "natively" (Unitree's in-house gait system ≠ industrial AMR dispatch) — you must pick one: ① bolt on SRC-5000 and let SEER be the main brain (recommended, least effort); or ② on the EDU/B2 Orin, self-write middleware translating state into VDA5050/custom protocol up to M4.
  • Which as the child dog: GO2 EDU = a low-cost closed-loop validation (≤7 kg light load, stair-climb); B2 = the heavy-load mainstay once cargo truly hits the 40 kg class. Consistent with the GO2 starter and haulage selection articles: choose by per-item cargo weight.

📦 Marsupial-shuttle warehouse transport: how to build it on SEER's system

Assembling the four blocks above into a "marsupial shuttle" — mother car carries the child dog across zones; the dog dismounts for the final stair-climb/squeeze-through leg — all under M4 dispatch:

Form How to connect Pros Risk / prerequisite
① all-SRC-native (least hassle) Mother car (SEER AMR or a third-party car with SRC) + child dog (with SRC-5000) both speak SEER-native; M4 tasks all One ecosystem, no protocol translation, plug-and-play If the mother car is Otsuka's existing vehicle, it likely needs an SRC swap/retrofit — conversion cost
② VDA5050 bridge (keep existing fleet) M4 as unified RCS: child dog runs SRC-native + Otsuka's existing AGVs run VDA5050, mixed dispatch Preserves the existing investment, no vehicle swap Prerequisite: Otsuka's existing AGV vendor exposes a VDA5050 interface — the biggest variable, check per-unit; and SEER's VDA5050 support is NOT stated on product home pages, only on solution/media pages — verify with SEER
③ self-integrated dock mechanism Either way, the physical "car stops → dog on/off → dock precision" mechanism (AprilTag/rail/ramp) has no off-the-shelf part This is precisely the in-house differentiation space Must be self-developed; dock precision and cross-floor return are the engineering hard parts (see integration-cases article)

Most realistic landing form: M4 as the unified RCS, the child dog (B2/GO2-EDU + SRC-5000) running SRC-native, and Otsuka's existing AGVs running a VDA5050 bridge — exactly "in-house software × off-the-shelf hardware + keep the existing fleet." But that VDA5050 path only holds if Otsuka's existing AGVs truly support VDA5050; otherwise either retrofit SRC onto the mother car too, or self-write middleware.

⚠️ Three honest gates (don't oversell)

  1. VDA5050 depends on Otsuka's existing AGVs: the standard itself targets only 2D wheeled AMRs, not quadrupeds; SEER's VDA5050 support isn't stated on product home pages — check both ends: do Otsuka's cars support it, does SEER's bridge cover that exact model.
  2. SEER has no local agent in Japan: only a German subsidiary; the JP site's inquiry routes to HQ. Landing requires direct contact with Shanghai HQ / Germany, or Otsuka doing Japan local integration & support (which is itself the cooperation entry point).
  3. GO2 light-load / B2 heavy-load: GO2 is rated ~7 kg, fit only for a light-load PoC; don't commit to a model before cargo weight is settled — first measure that key variable, per-item cargo weight.

🎯 Landing recommendations for Otsuka

  1. Step 1 (zero cost): have the JP team enter seer-robotics.ai/jp/download, pull the M4 catalogue and SRC manual, verify the first-hand wording on "third-party protocol onboarding / VDA5050 / quadruped interface," and replace media wording with evidence.
  2. Step 2 (PoC): GO2-EDU + SRC-5000 into M4, run the minimal loop "mother car carries dog → dog dismounts and climbs stairs → delivers → returns," validating dispatch orchestration and the marsupial dock.
  3. Step 3 (upgrade by cargo weight): once per-item weight is set, keep GO2 for light load, switch to B2 at ≥40 kg; for the mother car, choose "retrofit SRC" vs "VDA5050 bridge" by whether it supports VDA5050.
  4. Differentiation moat: the dock mechanism + cross "dog+AGV" dispatch orchestration + Japan local integration/support — none of these is an off-the-shelf shelf product industry-wide, exactly the "software × hardware" barrier in the business plan.

⚠️ To verify / follow-up

  1. Named compatibility list (logged as follow-up): investigate exactly which vendors' dogs / AMRs / AGVs can fit SEER M4 — a named list (which third-party cars have embedded SRC, which quadruped brands connected to M4 at WRC/CeMAT). Public material only says "quadruped dogs" without naming Unitree etc.; verify directly with SEER.
  2. M4's VDA5050 support: check the product manual / with SEER whether it covers Otsuka's specific existing AGV models.
  3. Whether the M4 product UI truly ships Japanese: verify by install or by asking SEER.
  4. Concrete downloads-center document list: enter via browser and fill in the M4 catalogue / SRC manual PDF titles and versions.