After 100 hours: what open-source humanoid hardware lets tinkerers actually experiment with

#robotics #opensource #humanoid

Menlo Research built walking humanoid legs in 100 days for under $30,000 in R&D — and open-sourced everything. Here is what that actually unlocks.

The problem open-source hardware solves

Humanoid robots are usually closed systems. You cannot inspect the control stack, swap actuators without a vendor relationship, or modify the locomotion policy without signing an NDA. For researchers and builders who want to study how these machines actually work, that is a hard stop.

Menlo Research's Asimov takes a different position. The mechanical designs, simulation models, and control algorithms are open. The "Here Be Dragons Edition" DIY kit — available for pre-order since Mar 2026 at $15,000 — is priced close to bill-of-materials cost. The company states that low-volume manufacturing should come in under $20,000 for the full body.

The question worth asking: once you have built it, what can you actually do with it that you could not do before?

What you are building

The Asimov v1 robot stands 1.20 m tall, weighs approximately 35 kg, and has 25+2 degrees of freedom (DoF) — the additional two DoF come from articulated toes. It uses a modular architecture: legs, torso, arms, and head snap together via universal motor mounting fixtures, so subsystems can be developed and tested independently.

The leg subsystem costs just over $10,000 all-in — roughly $8,500 for actuators and joint parts, the remainder for batteries and control modules. Most structural parts are compatible with Multi Jet Fusion (MJF) 3D printing, which produces functional parts without custom tooling. Teams without industrial MJF access can use alternatives including casting or standard 3D printing for most components. The knee plate — where stiffness and alignment are most critical — requires CNC machining, but Menlo redesigned it specifically to simplify that requirement.

Assembly takes approximately 100 hours, per Menlo's own account of building the legs subsystem.

What the architecture opens up

Locomotion research without lab access. The ankle design uses a parallel RSU (Revolute–Spherical–Universal) architecture rather than a simple serial joint. This gives two DoF — roll and pitch — with torque sharing between two motors, proximal placement of heavy components, and better backdrivability, meaning the ankle responds more naturally to ground contact forces. The design choices are documented, which means a tinkerer can study why they were made, test modifications, and observe the effects on gait directly.

Sim-to-real experimentation. Menlo built a parallel in-the-loop (PIL) simulator with intentional latency injection to match real hardware behavior. This is the gap where most academic research stalls: a control policy trained in simulation behaves differently on physical hardware because the simulator was too clean. With a documented PIL setup and the physical robot to test against, a builder can iterate on the sim-to-real transfer problem directly — not just read about it.

Gait tuning on your own hardware. Commercial robots ship with locked locomotion stacks. Asimov ships with an open control policy. That means a tinkerer can modify gait parameters, test different walking strategies, and measure the results on physical hardware — not just in simulation. For a researcher interested in locomotion, that is the difference between studying the subject and working on it.

Custom sensor integration. The modular architecture and open mechanical designs make it feasible to add sensors — cameras, force-torque sensors, tactile arrays — without voiding a warranty or waiting for vendor support. Each module has clear interfaces and documented load paths.

Manipulation research on the full stack. Different labs have different priorities. The modular design lets a team working on arm manipulation use just the torso and arms, while a team focused on locomotion works with the legs independently. A tinkerer building the full robot can switch focus without rebuilding the hardware.

What we know / what we don't

Verified:

Kit price: $15,000, close to BOM cost per Menlo Research
Full robot target: approximately $30,000 assembled
Specifications: 1.20 m tall, ~35 kg, 25+2 DoF
Leg subsystem cost: just over $10,000 all-in
Build time for leg subsystem: under 100 days at R&D pace
Manufacturing approach: MJF 3D printing for most structural parts, CNC for knee plate
Ankle design: parallel RSU architecture, documented
Pre-order available since Mar 2026, $499 deposit

Not verified / unknown:

Real-world build time for a tinkerer working alone (100 days was a team R&D effort)
Actuator pricing from Encos (vendor does not publish; requires direct quote)
How many kits have shipped or are in customer hands as of Apr 2026
Long-term part availability and community ecosystem maturity

What to watch

The Asimov kit is early. "Here Be Dragons" is not a product disclaimer to ignore. The questions that will determine whether this becomes a meaningful research platform:

Does a functioning community form around shared modifications, gait improvements, and sensor integrations?
Can builders outside well-resourced labs actually source and assemble the hardware at the stated cost?
How does the open locomotion stack perform compared to proprietary systems on standard benchmarks?

The modular architecture and open software stack are the right foundations. Whether the kit delivers on its research premise depends on what the first builders do with it — and whether they share what they find.