Yeah, when the G1s dance with kids, I realized how small they actually were. Definitely not the same category.

The H2 is what everyone is talking about now

some specs here: https://www.unitree.com/H2

claimed 3h battery life, can hold about 10% of its weight (7kg, with arms)

So Atlas can lift 7x the capacity. Even Digit, the tote-consolidating robot, can do 35lbs.

Unitree's demos are a lot of fun, and the antics of releasing the G1 to the public has certainly captured people's attention, but a "working" robot won't look, act, or develop from the G1 or even H2.

Unitree has plenty other industrial robots. https://www.unitree.com/ -> Click Robots

I wasn't trying to say that Unitree is somehow deficient. I'm sure they could build Atlas if they wanted.

My point was that BD could probably build a robot with the shown acrobatic capabilities, but they choose not to because their goal is to build robots that carry heavy loads for industrial applications.

They also wouldn't be getting any funding for doing such fun demos, even if they wanted to.

I don't need a robot to lift more than 15lbs in order to do all my maid work

Focussing on load capacity is missing the forest for the trees

The point is that a robot with higher load capacity is necessarily less agile. BD's target market is industrial so their robots are necessarily larger and less agile.

The fact that Unitree's robots appear so acrobatic reflects that they are likely on par with BD in terms of capabilities but have a different target market.

Isn’t it easier to make an agile robot big than to make a big robot agile?

One big issue is the "joints". It's always a trade-off between mass, strength, speed, precision, and dexterity. State of the art is matching or exceeding organic joints on 2-3 axis, while being an order of magnitude below human/animal performance on the two other axis.

Human individual fingers can withstand internal loads of hundreds of newtons, (possibly a thousand for brief periods if you're a star rock-climber), while at the same being capable of tasks such as writing, which require high speeds and (sub-)millimeter precision, while also enjoying 4 degrees of freedom (5 for the thumb), and they're stuffed full of sensory organs to boot. Oh and they're also self-healing, so taking some damage during use is no problem and they will actually adapt to the task they're used for over time. Everything we can make compared to this is laughably primitive.

If you make an agile robot big, it now weighs more, which means its joints now need to handle greater loads, which makes it necessary to make them much bulkier and heavier.

A lot of this is just inherent limitations of electric motors and having to convert rotational energy. This gets heavy fast.

Decent synthetic electrically-driven muscle fibers would go a long way.