Girls can't do math

Meanwhile, over at X.

We can see a pair of extensor tendons in the new 1X
Neo Gamma humanoid. Likely loops back to form pair of flexors underneath.

When tension is equal, finger extension/flexion. When differential, finger ab/adduction.

Pairing doubles grip strength. Fewer actuators, too.

Figure AI must be well capitalized if they can afford eggs for a demonstration. And from Whole Foods of all places.

Choices were made.

Looks like I will be using MapQuest going forward.

Bye, bye, google maps.

So this is why we are mad at Canada.

Amazon has been aggressively automating their warehouses, adding the famous Kiva robot at a rate of 1000 per day. Yes. 1000. Per. Day.

These Romba like bots fetch items for workers to pack into boxes.

This is the part Amazon has failed to automate even using state of the art industrial robots.

But it's not just analysts who believe that.

It is also bot oems. Both Agility and Apptronik are being used at GXO, the largest logistics company that you've never heard of.

First, who are the 2 customers? We know the 1st one is BMW. Who is the second?

Brett has said they are focused on two main customers in different industries.

This means the second is non-automotive.

The consensus is logistics, as most analysts see that as the largest and best opportunity.

Figure expects to deliver 100K humanoid bots in 4 years.

To just 2 customers.

How will they accomplish this?

A breakdown 🧵

Evidently Figure AI is shipping F.02 to customers in carbonite,

Wow! OneDrive created November memories for me.

About sums it up.

AI impresses me more and more each day.

Despite this software challenge, again it will be a test of hardware design.

The business-end of a robot are its hands. They need to be durable, yet dexterous enough to position the hand and fingers to separate parts, and grasp them.

Finger and wrist design is crucial.

While placement accuracy is now very reliable, with error correction if necessary, the real test will be the pick.

Once the part is in hand, the placement is very structured.

Unlike this demonstration, what is not structured is the picking.

Parts are usually a jumble inside of a parts bin.

Only after achieving reliability, is it time to increase the speed until something breaks. Usually the software. Then improvements are made, and the dial can be turned up again.

Figure has increased the speed 4X, but not yet to human speed. The actuators still have headroom. It will get faster.

Go Figure

What's the big deal about 1000 placements/day?

Reliability.

Before humanoid bots can be deployed they must prove reliability both in task completion and durability.

One is a S/W issue, the other H/W. 🧵