Humanoid Robots for Moon and Mars

[Twark_Main]

Power will be an ongoing issue though, as you will need an some battery exchange infrastructure and a large inventory of charging batteries.

Tesla has already amply demonstrated that this "need" is anything but.

If anything, that's the lowest-risk part of the whole project.

[lamontagne]

To improve:

1) delete the legs. Use wheels (or possibly treads).
2) redo the arms mechanically.
3) Move all actuators to the torso
4) make as much as possible cable-driven or hydraulic.
5) high pressure hydraulics to reduce weight and inertia.
6) reduce hands to 3 digits.

So you should like my robot :-)

[lamontagne]

Not certain about the hydraulics for Teslabot.  Musk clearly stated less powerful than a human, so electric actuators seem like a safer solution.
I believe these are already implemented in the new classes of "soft" or 'smart" robot arms for interoperations with humans. 

If you want a lot of power, use a dumber but stronger additional piece of equipment that the robot operates.  Don't build overcapacity into the robot.  You want a physically weak but relatively smart robot.

[Robotbeat]

You need a lot of power just to have it walking around carrying its own mass without accidentally toppling over or being extremely slow. Like nearly every bipedal robot before Boston Dynamics.

[edzieba]

Boston Dynamics' Atlas is an existence proof that an electromechanical (not electrohydraulic) actuator driven bipedal humanoid is both feasible and possible, and able to carry practical loads. And its use by multiple teams in the DARPA Robotics Challenge is also a demonstration that the software to task and direct the robot autonomously remains a challenge even if that of building and manipulating the robot is a solved one.
Or in other words, the problem of having a robot walk up to a drill (locomotion challenge), pick it up (object recognition, characterisation, and manipulation challenge), and drill a hole in a wall (force-based manipulation challenge) is a solved one. The problem of having a robot do so by telling it "go drill a hole here" (tasking challenge) rather than an explicit canned task sequence with a few token variables sprinkled in is an unsolved one.

[Robotbeat]

Nope. Atlas is electrohydraulic.

[billh]

Engineered Arts is using electric motors of their own design, and I am fascinated by how smooth the motion is. Even the most sophisticated robots generally seem to have noticeable slack and hysteresis in their movements. It often interferes with things like the fine control needed to, say, pick up or put down a container. However, I'm guessing EA's motors are not very powerful. I wonder if Ameca could pick up even 1 kg.

[Cheapchips]

Thought this paper from MIT was interesting in the context of electric actuator performance. They emulated Atlas's somersault/backflip using scaled up versions of Mini Cheetah's open source actuators.  While it was a simulation, they did build the actuators and endured that they'd perform off their chosen battery.

Teslabot obviously doesn't need to be doing backflips to be (maybe) useful.  They probably dont need actuators that powerful?

[Robotbeat]

You need powerful actuators to do useful work and to be stable without falling over.

[Cheapchips]

You need powerful actuators to do useful work and to be stable without falling over.

How much power do you think it going to use to stand? It's not constant max power.  It's a sequence of fine balancing wibbles, for want of a better term.

Humans use 11.6 Wh to stand  No idea if evolution did a mechanically good or bad job on that mind. 

[Robotbeat]

Very little to stand. But to be stable while walking, you need substantial torque and speed in your actuators.

The average power is low. Peak power high.

[Twark_Main]

Humans use 11.6 Wh to stand

Warning: bad units detected.

You're giving a quantity with units of energy (11.6 watt hours, ie 41,760 joules). But for this metric, we should expect a quantity with units of power (watts).

Did you mean 11.6 watts?

[Cheapchips]

Probably an ignorant conversion on my part? Humans burn 100 calories an hour to stand.

[Twark_Main]

Probably an ignorant conversion on my part? Humans burn 100 calories an hour to stand.

Thanks! 116.3 watts, then.

If you subtract out a basal metabolism of 2000 kilocalories per day, that's 19.4 watts.

If you subtract out a basal metabolism of 2500 kilocalories per day, that's -4.8 watts.

I don't know quite what to make of this.

[edzieba]

Very little to stand. But to be stable while walking, you need substantial torque and speed in your actuators.

Well, no. Passive-dynamic walkers can stably walk with no input actuator power nor any active control at all, effectively rolling down an inclined plane:

Note the swinging knee-locking gait, and lack of static balance (i.e. if you stopped the walker at any point in its walk cycle it would topple). This is as opposed to older concepts like Honda'a Asimo, which used a bent-leg gait and continuous active stabilisation (at any point in its walk cycle it can halt in place and remain balanced). Modern waling robots aim to use a passive-dynamic cycle to reduce power requirements, with actuators modifying the gait on the fly for steering, obstacle tolerance, and start/stop.
You can see the development of this in Boston Dynamics' quadruped work, where the original 'Big Dog' maintained balance through near constant leg movement (the distinctive stationary shuffle) to separate the start/stop cycle from actual acceleration (first start walking in place to move to dynamic balance mode, then start moving forward). Moving on to today's Spot, they can now go from a standing start to a walk or run without this intermediary stage.

[Robotbeat]

Very little to stand. But to be stable while walking, you need substantial torque and speed in your actuators.

Well, no. Passive-dynamic walkers can stably walk with no input actuator power nor any active control at all, effectively rolling down an inclined plane:

Note the swinging knee-locking gait, and lack of static balance (i.e. if you stopped the walker at any point in its walk cycle it would topple). This is as opposed to older concepts like Honda'a Asimo, which used a bent-leg gait and continuous active stabilisation (at any point in its walk cycle it can halt in place and remain balanced). Modern waling robots aim to use a passive-dynamic cycle to reduce power requirements, with actuators modifying the gait on the fly for steering, obstacle tolerance, and start/stop.
You can see the development of this in Boston Dynamics' quadruped work, where the original 'Big Dog' maintained balance through near constant leg movement (the distinctive stationary shuffle) to separate the start/stop cycle from actual acceleration (first start walking in place to move to dynamic balance mode, then start moving forward). Moving on to today's Spot, they can now go from a standing start to a walk or run without this intermediary stage.

What they didn’t show you is all the times it fell over. That’s the real key.

[RonM]

Probably an ignorant conversion on my part? Humans burn 100 calories an hour to stand.

Thanks! 116.3 watts, then.

If you subtract out a basal metabolism of 2000 kilocalories per day, that's 19.4 watts.

If you subtract out a basal metabolism of 2500 kilocalories per day, that's -4.8 watts.

I don't know quite what to make of this.

What strange kind of math are you people trying to do here?

[Jim]

Somewhat reluctant to start this thread, as it's all really early, but then Moon and Mars missions are 3+ years away...

At Tesla's AI day they announced they are working on Tesla Bot, leveraging the work Tesla already does with computer vision, robotics and volume manufacturing.  A "we might as well do it, as we have all the pieces" move.

This thread is just another case that everything is a nail and the hammer is SpaceX/Tesla.

[lamontagne]

Probably an ignorant conversion on my part? Humans burn 100 calories an hour to stand.

Thanks! 116.3 watts, then.

If you subtract out a basal metabolism of 2000 kilocalories per day, that's 19.4 watts.

If you subtract out a basal metabolism of 2500 kilocalories per day, that's -4.8 watts.

I don't know quite what to make of this.

What strange kind of math are you people trying to do here?

The usual mixup between energy and power, and using one for the other.  Plus averages vs peak uses.
2400 calories (average daily food intake) is 100 calories per hour on average.  However we use more when moving and less when resting.
And it would seem 100 calories per hour when standing.  That's really 100 kilocalories per hour, or 418400 j/h = 116.2 J/s, or 116 watts.  Instant demand, to determine electric/hydraulic motor power, may be higher.

When I'm pulling hard on the rowing machine I can get 300+ watts for a few seconds at the gym :-)  but it's a lot of work!

[Robotbeat]

It’s worth keeping in mind that power is kind of missing the point… electric motors have excellent power to weight ratio when spinning very fast. And TERRIBLE power to weight ratio when moving slowly. So for the same task, an electric motor requires a lot more energy and will be heavier than you think. To compensate, typically very high gear ratios are used, often hundreds to one, and that reduces efficiency, increases mass and complexity, and makes the motors not really very backdrivable (and therefore not as responsive to stimuli).

BLDC motors are getting good enough that 10:1 planetary gearboxes, which are a lot more backdrivable, are feasible-ish for smaller robots (used on the MIT Mini-Cheetah), but you’re still operating with poorer performance than you’d like. The human body is a mechanically very efficient machine with actuators very well optimized for the speed range we typically operate in. Going slower doesn’t help much (if any) because you’re already operating at the lower speed end of your motor’s operating range, even with 10:1 gearing. Using higher gearing has the drawbacks I already mentioned.

It’s a hard problem. (And I don’t think Tesla bots will be terribly relevant for Mars except in niche cases, like with the NASA humanoid robot. You’d be better off with some sort of optimized non-humanoid robot, most likely.)