Humanoids vs traditional robotics: when a humanoid makes sense and when a cobot or AMR is better

"Can I replace the cobot on the line with a G1?" Almost always the answer is no. The humanoid works, but it solves a different problem. This guide compares the two families using the specifications declared by the manufacturers, the public list prices and the current state of the safety standards, to help you decide where to start.

What we mean by traditional robotics

• Cobots: collaborative arms for screwdriving, pick and place, machine tending and palletising. Example: the Fairino FR3, 6 axes, 3 kg nominal payload, 622 mm reach, ±0.02 mm repeatability according to ISO 9283 (manufacturer datasheet). Full range on the cobots page.
• AMRs: autonomous mobile robots for transporting carts, racks and pallets on indoor routes. At list price, the range starts with the AutoXing Juno Plus. Overview on the AMR page.
• Quadrupeds: inspection, surveillance and sensor transport in unstructured environments, with models such as Go2, AS2 and B2.

What they have in common: defined cycles, established reference standards and KPI that are easy to measure, such as parts per hour, OEE and cycle time.

What a humanoid brings today (and what it doesn't)

A biped such as the G1 or the H2 moves through spaces designed for people, can switch tasks within the same day and is the platform on which research on embodied AI, teleoperation and imitation learning is done today. The official G1 specifications help to understand the limits:

Unitree notes that the arm payload varies greatly with the extension posture and does not declare a repeatability value. Realistic uses today:

• university laboratories and research projects;
• PoC in a fenced area: light kitting, data collection, teleoperation;
• technology demonstrators and internal training.

It more rarely makes sense to replace an already efficient cell with a humanoid.

Direct comparison (indicative values)

Prices are updated on the Unitree price list and on the product pages. On top of the humanoid's price you should add software development, an equipped test area and training: these often weigh as much as the hardware.

Safety and standards

General information, not legal advice.

The ISO 10218-1:2025 and ISO 10218-2:2025, published in February 2025, replace the 2011 editions and cover industrial robots, applications and cells. Part 1 now incorporates the requirements for collaborative applications that were previously in the ISO/TS 15066 technical specification.

A biped adds one more problem: it stays upright only thanks to active balance control. If it loses power or is brought to a stop it can fall, whereas an AMR or a cobot simply stops where it is. For this category (industrial mobile robots with actively controlled stability, bipeds and quadrupeds) the ISO is working on ISO 25785-1, not yet published.

From 20 January 2027 the Machinery Regulation (EU) 2023/1230 applies, replacing the Machinery Directive 2006/42/EC. What it means for Unitree robots: Machinery Regulation 2027. In practice, the first tests with a humanoid are carried out in a fenced area, with a fall-prevention system (harness or overhead crane) and operators trained on the stop procedures.

When to stick with traditional robotics

• High-volume tasks with a short cycle.
• You need a machine with CE marking and quick commissioning.
• The environment is already structured: conveyors, bays, defined routes.
• The investment must pay back with measurable savings in the short term.

When to consider a humanoid

• Tasks not yet formalised, or many different low-volume tasks.
• R&D project, university, research grant.
• An environment designed for people, hard to adapt with conveyors or AMRs.
• A budget dedicated to innovation and data collection, not to the direct replacement of labour.

There is also a middle ground: the G1-D is a legless dual-arm torso, on a fixed base (G1-D Standard, from €38,005) or on a mobile base (G1-D Flagship, from €51,944). It removes the risk of the biped falling and stays closer to a manipulation cell. To choose between models: which Unitree humanoid to choose.

The path we usually recommend

1. First: automate the bottleneck with a cobot or AMR, where the return is easier to measure.
2. Then: a humanoid PoC on a complementary task (inspection, replenishment, data collection), with KPI written down before you start.
3. Scale-up: only if the PoC KPI hold up (success rate, cycle time, total cost).

For a company purchase, cobots, AMRs and humanoids may qualify for several UK schemes that reduce the net cost, from capital allowances to R&D tax relief: the details are on the UK incentives page. If you are not sure where to start, the free assessment is there for exactly this: Abra Robotics supplies the Unitree range also in the United Kingdom and Ireland, with support in English, but it also sells cobots and AMRs, so there is no reason to push a humanoid where it is not needed.