"Atlas’ current dance performance uses a mixture of what we like to call reflexive control, which is a combination of reacting to forces, online and offline trajectory optimization, and model predictive control. We leverage these techniques because they’re a reliable way of unlocking really high performance stuff, and we understand how to wield these tools really well. We haven’t found the end of the road in terms of what we can do with them"
"offline trajectory optimization" this doesn't sound good... I read this as : "we've made those robots do things they won't be able to do autonomously thanks to massive amounts of offline optimization".
Sure it's impressive, but I wonder how it helps the casual viewer to get an idea of what these robots can actually do. One could say that's the whole point of PR but...
It sure helped BD engineer to push their robots forward (as said in the article).
And it doesn't matter that the robot couldn't possibly coordinate everything in the video on is own. The software doesn't just have to coordinate high level "go here, move like this", you also have the lower level coordination that sits between the collision avoidance and basic driving etc and the $gcode_variant driving the servos. I'm guessing that stack gets a major overhaul every time one of these videos gets made, like almost every Hollywood movie has a massive "software development" section because of all the artistic direction that needed new or changed functionality.
The original video showing Spot moonwalking had the most imperceptible of splices in it right at the end of the moonwalking bit. Several months' worth of development alongside careful use of multiple scenes in this newer video have mitigated the need for hacks like that. It's nice to see.
Personally, I like seeing these baby steps and starts. As you and the article pointed out, this initiative let them see what they need to do before they can get to a purely autonomous dancing robot. Exciting times
"Atlas’ current dance performance uses a mixture of what we like to call reflexive control, which is a combination of reacting to forces, online and offline trajectory optimization, and model predictive control. We leverage these techniques because they’re a reliable way of unlocking really high performance stuff, and we understand how to wield these tools really well. We haven’t found the end of the road in terms of what we can do with them"