Robo-dog was the scare a few months back.

Now it's the Atlas robot they built. The running is so-so but that jump was impressive.

Yeah, the humanoid acrobatics are very impressive... They've made a helluva lot of progress since the first teetering/wobbly/unbalanced steps that I remember...

But I thought I'd start this thread with the doggie-bots... perhaps because it's so easy to imagine a real-world use for them: maybe as "watchdogs" that patrol a property at night, beaming security video back to property owners or law enforcement... etc. etc.

perhaps because it's so easy to imagine a real-world use for them:

You could equip them with a flamethrower and 20 gallons of napalm.

That would work too.

Well yeah... I'm sure there are dozens of ways to weaponize the little contraptions.... The "scarey" part is the "artificial intelligence".... the programming & processing power that makes it seem like they're entirely "thinking" all on their own...

Well yeah... I'm sure there are dozens of ways to weaponize the little contraptions.... The "scarey" part is the "artificial intelligence".... the programming & processing power that makes it seem like they're entirely "thinking" all on their own...

They aren't all that smart.

Look, for instance, at the Atlas robot running. Notice how the feet are placed on each step. In order to keep the robot from tripping over its own "toes", they don't lift the foot off the ground in a fluid forward motion as a human does. Instead, they actually move the foot backward (relative to the robot body) as they lift the foot off the ground. It's not a natural or efficient motion at all. Even very young children can run better than that.

If they could overcome that problem, they could make a robot that could run like a human being and probably double or triple the speed of its running.

Just making a few observations on the engineering and the state of the art in android software.

Similarly, most of the acrobatics like the log jump or the backflip are set pieces and can only be performed on a very level surface and only within certain parameters. They haven't really solved the generalization problems in processing their sensor data, especially the visual/lidar sensors.

It's clever enough but it's no Skynet either. Not yet.