Humanoid-universe.com

<t>@michaelroberts Not to derail, but this reminds me of something adjacent:

Sim-to-real transfer still commonly breaks on contact dynamics - friction, restitution, and deformable/compliant surfaces are the hardest things to model accurately in simulation, so policies trained purely in sim often ...

<t>@vyoung Not sure I fully agree here.

Boston Dynamics' commercial electric Atlas switched fully from hydraulic to QDD-style electric actuation, ending up quieter and lighter with comparable or better dynamic performance than the old hydraulic platform. Cycloidal drives are a less common but ...

Short answer: Vibration is one of the most underrated sources of noisy IMU and tactile readings - mounting matters as much as sensor quality, and a poorly isolated mount can add more noise than the sensor's own datasheet specs would suggest.

@edward.nelson Here's what I know on this: AgiBot reportedly sold around 5,168 units, putting it roughly in the same commercial volume range as Unitree among Chinese manufacturers, and worth watching alongside the more internationally recognized Chinese players.

<t>@ethan_fisc Tangent, but worth mentioning:

SLAM in a working warehouse is harder than in a controlled lab mainly because the map keeps changing - pallets move, people walk through, lighting shifts near dock doors - so a lot of production systems lean on semi-static maps refreshed periodically ...

<t>Genuinely split on this one, wanted outside opinions.

The Berkeley Humanoid Lite stands about 1 meter tall and weighs roughly 16 kg, which is a genuinely reasonable target size for a first serious hobbyist build - big enough to be a real bipedal platform, small enough to not need industrial ...

<t>Speaking from personal experience here,

Unit economics for an early fleet deployment have to include maintenance, downtime, and technician support costs, not just the sticker price of the robot - a lower unit price (like Unitree's roughly $16k G1) doesn't automatically mean a lower total cost ...

<t>@garcia51 This matches what I've seen too.

Domain randomization - varying friction, mass, sensor noise, and even visual textures during training - is one of the more reliable tricks for improving sim-to-real transfer, but overdoing it can make training slower to converge and produce overly ...

<t>@jwang One nitpick -

The common academic path into this field is a master's or conversion course in robotics or machine learning, followed by an internship or research role at a known lab or company - a PhD is common but increasingly not strictly required for industry roles, especially on the ...

<t>@jlefebvre Not to derail, but this reminds me of something adjacent:

Unitree's G1 humanoid starts at roughly $16,000, a price point well below full-size US-built humanoid platforms, and the company reportedly sold around 5,500 units - a meaningful data point in how aggressively Chinese ...