Interesting post. Atlas could maybe be called a "post-humanoid" robot: a design that has evolved beyond straightforward imitation of the human form, while retaining the strengths of it.
A question about one aspect of post-humanoid robotic designs: sometimes we've seen robots that put wheels on their legs (instead of feet), so they can switch between walking and rolling depending on the situation. That seems to not be in favor recently, at least not with humanoids. Do you know why wheels-on-legs isn't popular?
Thanks, Chris, for the well-argued and written post, as usual. It well-articulated some of my developing thoughts from earlier this week on the implications of a human-demonstration-based pretraining architecture on the robot form itself. I think in addition to the limb morphology, there are similar implications on sensor layout for behaviors that need sensory feedback, such as sparse foothold mobility and tricky manipulation.
Interesting, but I will push back on one minor point you make. I think you're wrong about dogs and whales not using their "fingers." Your assumption that digits are only useful if used as humans use them, but that's a bit speciesist. A whale and a dog would almost certainly move differently if they didn't have those digits.
When I see Atlas I 💯 think “Oh that looks like a human”, it just has “human-plus” movement capabilities. Great article overall! My main takeaway for a that lots of companies will probably try lots of design strategies and I am very excited to see what they come up with.
This is excellently reasoned. Your point about economical design cutting through the mythology of human-mimicry is spot-on - the only two unique actuators thing is genius from a supply chain perspective. I worked on manufacturing optimization years ago and watching companies contort themselves to reduce SKU count was always fascinating. The argument that robots need to use human tools feels particuarly weak when you think about how quickly infrastructure adapted to cars. I dunno why anyone would assume tools wont evolve similarily. Really insightful analysis of the Atlas design philosphy and what it signals about practical robot deployment versus marketing-driven anthropomorphism.
Interesting post. Atlas could maybe be called a "post-humanoid" robot: a design that has evolved beyond straightforward imitation of the human form, while retaining the strengths of it.
A question about one aspect of post-humanoid robotic designs: sometimes we've seen robots that put wheels on their legs (instead of feet), so they can switch between walking and rolling depending on the situation. That seems to not be in favor recently, at least not with humanoids. Do you know why wheels-on-legs isn't popular?
Thanks, Chris, for the well-argued and written post, as usual. It well-articulated some of my developing thoughts from earlier this week on the implications of a human-demonstration-based pretraining architecture on the robot form itself. I think in addition to the limb morphology, there are similar implications on sensor layout for behaviors that need sensory feedback, such as sparse foothold mobility and tricky manipulation.
Yeah I imagine that for example these robots will have many more "eyes" than a human. This is certainly true of the Figure 03!
Interesting, but I will push back on one minor point you make. I think you're wrong about dogs and whales not using their "fingers." Your assumption that digits are only useful if used as humans use them, but that's a bit speciesist. A whale and a dog would almost certainly move differently if they didn't have those digits.
When I see Atlas I 💯 think “Oh that looks like a human”, it just has “human-plus” movement capabilities. Great article overall! My main takeaway for a that lots of companies will probably try lots of design strategies and I am very excited to see what they come up with.
This is excellently reasoned. Your point about economical design cutting through the mythology of human-mimicry is spot-on - the only two unique actuators thing is genius from a supply chain perspective. I worked on manufacturing optimization years ago and watching companies contort themselves to reduce SKU count was always fascinating. The argument that robots need to use human tools feels particuarly weak when you think about how quickly infrastructure adapted to cars. I dunno why anyone would assume tools wont evolve similarily. Really insightful analysis of the Atlas design philosphy and what it signals about practical robot deployment versus marketing-driven anthropomorphism.