Robot buyers face a category that now stretches from a 15 kg quadruped to a 70 kg humanoid, yet the spec sheets do not describe one common job. Unitree and Booster Robotics divide the market by mobility, load, computing access, protection, and development rights. South Korean procurement records add a second view: public bodies are spending across humanoids, quadrupeds, surgical systems, underwater walkers, cleaning robots, and other robot types, but the records do not prove that any named product in this comparison won those contracts. The useful comparison is therefore not a single winner. It is a map of what each design gives up to reach a price, weight, or operating target, followed by a clear boundary between manufacturer claims and public buying data.
What the spec sheets show
Unitree's Go2 is the lightest quadruped in the packet at about 15 kg with its battery. Its standing body measures 70 cm by 31 cm by 40 cm, and all trims list 12 joint motors and a 4D LiDAR L2 with a 360° by 96° field. The trim ladder changes performance. Air reaches 2.5 m/s, Pro reaches 3.5 m/s, and EDU reaches 3.7 m/s, with a stated maximum near 5 m/s. Payload also rises from about 7 kg on Air to about 8 kg on Pro and EDU, while stated maximum payload moves from about 10 kg to about 12 kg on EDU. The Unitree Go2 specification page puts secondary development, an Orin compute module, and a depth camera on EDU only. That makes software access, rather than basic sensing, the clearest trim boundary.
The A2 changes the use case through mass, endurance, and environmental protection. It weighs about 42 kg with a battery and about 35 kg without one, compared with Go2's roughly 15 kg. Unitree lists a continuous walking load of about 25 kg, an ideal figure of about 35 kg, and a standing load of about 100 kg. The machine runs unloaded for more than 5 hours or about 20 km, while a 25 kg load lowers those claims to more than 3 hours or about 12.5 km. Its speed remains 3.7 m/s, with a maximum near 5 m/s. The Unitree A2 specification page lists IP56 for A2 and IP56 to IP67 for A2-PRO, whose core components carry the IP67 rating. Its operating range is -20℃ to 55℃.
As2 sits between those quadrupeds. Every trim weighs about 20 kg, while torque runs from about 65 N·m on AIR to about 75 N·m on PRO and about 95 N·m on X and EDU. Continuous walking load rises in the same order, from about 10 kg to about 13 kg and then about 15 kg. Maximum standing load moves from about 45 kg to about 55 kg and about 65 kg. AIR has no published protection rating, while PRO, X, and EDU list IP54. Booster's T2 follows a different geometry. The Booster T2 specification page lists a humanoid about 1.4 m tall and about 42 kg, with 31 degrees of freedom, a maximum dual-arm payload of 10 kg, a 3 m/s walking speed, and 2 hours of continuous walking.
Where the designs diverge
The quadruped trade-off is easiest to see in load versus portability. Go2 can carry about 7 kg or 8 kg continuously depending on trim, while A2 lists about 25 kg and As2 ranges from about 10 kg to about 15 kg. A2 pays for that capacity at about 42 kg, almost the same listed mass as the T2 humanoid. As2 weighs about 20 kg, which keeps it closer to Go2 while offering as much as about 95 N·m of maximum joint torque on X and EDU. Go2's packet does not publish a maximum joint torque, so a direct torque comparison is not available. A2 publishes about 180 N·m. The sheets also separate slope claims: Go2 Air lists 30°, Go2 Pro and EDU list 40°, As2 ranges from about 30° to about 40°, and A2 lists about 45°.
The humanoids divide along scale and articulation. Unitree R1 stands 1,230 mm tall and weighs about 27 kg in Air form or about 29 kg as R1 and R1 EDU. Its total degrees of freedom start at 20 for Air, move to 26 for R1, and range from 26 to 40 for EDU. Unitree H2 stands 1,820 mm tall, weighs about 70 kg, and lists 31 degrees of freedom. Its arm joint torque reaches 120 N·m, leg joint torque reaches 360 N·m, and arm payload is rated at about 7 kg with a peak near 15 kg. T2 also lists 31 degrees of freedom, but its maximum peak torque is 140 N·m and maximum dual-arm payload is 10 kg. The packet does not give one common torque location or payload posture, so those values are not fully equivalent.
Compute and development access create another split that physical dimensions cannot show. T2 Education lists a high-performance ARM processor with 375 TOPS, while Professional lists a Jetson Thor T5000 with 2,070 TFLOPS. Those units differ, so the packet does not support a direct ratio. R1 uses an 8-core CPU as standard, and R1 EDU offers Orin modules rated at 40 to 100 TOPS. Go2 reserves secondary development, an Orin module, and a depth camera for EDU. H2 also limits secondary development to EDU. However, a larger compute figure does not establish better field performance by itself. Sensor inputs, power limits, software tools, and task success are not reported on a common test. The risk is reading processor labels as a finished autonomy score when the packet gives no shared benchmark, no shared workload, and no published result that compares these machines under the same operating conditions.
Price and timing reveal the category split
Published store prices place the low end far below the full-size humanoid. Go2 Air is listed at $1,600 and Go2 Pro at $2,800, excluding tax and shipping. As2 Air is listed at $3,100, while As2 Pro is $4,650. R1 Air is $4,900 and R1 is $5,900, also excluding tax and shipping. H2 rises to $29,900, excluding tax and shipping. The packet does not publish prices for A2, T2, Go2 EDU, As2 X, As2 EDU, R1 EDU, or H2 EDU. It also does not provide a consistent delivered price, option price, tax figure, or shipping cost. As a result, the published values show entry points rather than complete ownership costs. They still expose a broad ladder, with compact quadrupeds and a lightweight humanoid grouped well below the larger H2 body.
Announcement dates show how quickly manufacturers are filling adjacent sizes and roles. Unitree announced Go2 on 2023-07-12, R1 on 2025-07-25, A2 on 2025-08-05, and H2 on 2025-10-20. As2 followed on 2026-02-24. Booster announced T2 on 2026-07-13, according to the Booster T2 launch announcement. The sequence does not show a single move from quadrupeds to humanoids. Unitree added R1, A2, H2, and As2 around different mass and job requirements. The result is overlapping coverage: lightweight mobility in Go2, intermediate load in As2, industrial load and protection in A2, compact humanoid development in R1, and a full-size body in H2. T2 enters as another development humanoid with its own compute split.
The price sheets also diverge from the engineering sheets in what they omit. Go2's public entry prices cover Air and Pro, but the EDU trim contains the listed development features. As2 prices cover Air and Pro, while X and EDU carry the highest torque, the highest continuous walking load, the highest standing load, and the higher operating-temperature ceiling of 55℃. R1 prices cover Air and R1, while EDU alone offers the optional dexterous hand and the broader 26 to 40 degree-of-freedom range. H2 publishes a price for H2, while EDU carries secondary development and multiple dexterous-hand options. Buyers can compare the lower trims by sticker price, but the packet does not publish enough higher-trim prices to calculate the cost of software access, added compute, hands, protection, or maximum mechanical capability.
Hidden insight: public buying is broader than product rivalry
South Korean public procurement data covers 1,248 notices between 2025-11-03 and 2026-11-03, with a combined value of KRW 189,501,725,527. At the packet's KRW 1,342.9 per dollar rate dated 2026-11-02, that is about $141,113,802.61. The Korea ON-line E-Procurement System records are grouped into 194 humanoid notices worth KRW 21,882,789,320, about $16,295,174.12; 74 quadruped notices worth KRW 6,837,867,833, about $5,091,866.73; and 980 other robot notices worth KRW 160,781,068,374, about $119,726,761.76. The largest spending pool is therefore the broad other-robot segment, not the two forms described by the manufacturer comparison. These totals describe notices, not confirmed purchases of Unitree or Booster hardware.
The most active named public bodies are research and engineering organizations. The Korea Institute of Industrial Technology (한국생산기술연구원) appears in 126 notices. The Korea Advanced Institute of Science and Technology (한국과학기술원) appears in 53, the Korea Institute of Machinery and Materials (한국기계연구원) in 50, the Seoul National University R&DB Foundation (서울대학교산학협력단) in 41, and Ecopiece Co., Ltd. (에코피스 주식회사) in 31. Notice counts indicate activity, not spend by institution, because the packet does not provide institution-level amounts for this ranking. The buyer list fits research, testing, and applied engineering work, but the data does not publish which of the compared products, if any, each body selected. It also does not separate completed awards from every other notice status in the top-institution table, so counts should not be treated as installed units.
The largest listed notice is Daejeon Regional Public Procurement Service (조달청 대전지방조달청), “Production of a Completed Prototype Underwater Walking Robot” (수중보행로봇 완성시제 제작), dated 2026-09-10 and valued at KRW 4,097,604,960, about $3,051,310.57. Jeju National University Hospital (제주대학교병원) appears with “Purchase of One Set of a Robotic Surgical System, Medical Device, Reannouncement” ((의료기기) 로봇수술기 1SET 구매(재공고)) on 2026-02-20 for KRW 3,325,000,000, about $2,475,984.81. The same hospital title without the reannouncement label appears on 2026-02-10 and twice on 2026-02-09, each at the same amount. These repeated entries may represent procurement stages or repeated notices. The packet does not identify them as four separate delivered systems, so summing them as completed purchases would overstate the evidence.
Defense procurement adds small but concrete signals without naming a compared model. The packet lists a plan called “Purchase of Unit Equipment Cleaning Robots” ((26-물품-69) 부대비품 청소로봇 구매) for KRW 16,390,000, about $12,204.93, due in 2026-11. It also lists a result called “Rental Service for Multi-Legged Walking Robots for Combat Experiments” ('26년 전투실험용 다족보행로봇 임차용역), due 2026-09-08, and a plan called “AI Robot Application Demonstration Service” ((26G056-H)AI 로봇 적용 실증 용역), due in 2026-11. Amounts for the latter two are not published in the packet. The defense entries show interest in cleaning, multi-legged mobility, and applied testing, but they do not connect those requirements to Go2, A2, As2, T2, R1, or H2. That boundary matters when comparing public demand with vendor specifications.
What to watch next
The first gap to watch is publication of prices for the capability-rich trims. Go2 EDU contains the listed development path, while As2 X and EDU carry about 95 N·m torque and about 15 kg continuous walking load. R1 EDU reaches as many as 40 degrees of freedom and can take an optional dexterous hand. H2 EDU adds secondary development and multiple hand options. None has a published price in the packet. A2 and T2 also lack published prices. Until those figures appear, entry-price comparisons favor models whose base trims are listed, but they cannot show the price of a deployable research configuration. A complete comparison would need the exact trim, compute module, sensors, hands or grippers, protection level, tax, and shipping. Those values are not published here, so the current data supports a price ladder only for the named store configurations.
The second gap is common testing. Go2, A2, and As2 publish speed, load, slope, and mass, but they do not share a packet-level test protocol. T2, R1, and H2 publish degrees of freedom and different compute or torque details, yet the sheets do not report a shared manipulation task. A2's more than 5 hours unloaded and more than 3 hours with 25 kg cannot be compared directly with T2's 2 hours of continuous walking because body form, load, terrain, and test conditions differ. H2's roughly 3 hours on a 0.972 kWh battery also belongs to another body class. Watch for published test conditions, not just higher headline values. Without them, the packet shows design intent and claimed limits, while real task efficiency and reliability remain unpublished.
The third gap is procurement outcome data tied to exact models. Monthly notice counts rise from 42 in 2026-02 to 169 in 2026-08, then register 145 in 2026-09 and 163 in 2026-10. The partial 2026-11 count is 8 as of 2026-11-03. These figures show when notices appeared, but they do not identify awards, deliveries, model shares, or field results. The next useful data would link a notice to a selected model, final contract amount, delivery date, and operating role. For now, the safest conclusion is narrower: South Korean public bodies are active robot buyers, the largest money pool covers other robot systems, and manufacturer sheets separate products more clearly by load, body scale, protection, compute access, and development rights than by one universal performance score.
South Korea logged 1,248 public robot notices worth KRW 189,501,725,527, while the compared machines ranged from a 15 kg quadruped to a 70 kg humanoid.
Key Takeaways
- Go2 prioritizes low mass and low entry prices, while A2 trades about 42 kg of body mass for about 25 kg of continuous walking load and stronger protection.
- As2 fills the middle with about 20 kg mass, up to about 95 N·m joint torque, and up to about 15 kg continuous walking load.
- R1, T2, and H2 separate humanoid development by body size, articulation, compute options, torque, and access to secondary development.
- Published prices run from $1,600 for Go2 Air to $29,900 for H2, but several development and higher-performance trims have no published price.
- South Korean procurement covers 1,248 notices worth KRW 189,501,725,527, but the records do not establish purchases of the compared models.
Questions Worth Asking
- What do EDU and Professional configurations cost after compute, sensors, hands, tax, and shipping?
- Which machines can complete the same mobility or manipulation task under a shared test protocol?
- Which South Korean notices became awards, which exact models won, and how did the hardware perform after delivery?
Values come from manufacturer spec pages and public procurement records compiled by the TechFastForward data desk, linked inline.
