Successfully test the new version of Figure 2

The evolution of robotic learning.

Figure IA showed the world an update of its humanoid robot called Figure 2 and now it has just reached a milestone that could redefine all warehouse logistics and a seamless video published by them shows the robots working non-stop for an entire hour, organizing packages with an average time of just 4 seconds per item.

And one of the points that I found most interesting that it learns from its mistakes, adapts its grip and remembers what it saw seconds before, is the beginning of a new generation of really useful robots, until recently that same robot took 5 seconds to handle each package, now after just 4 months of progress it does the same work with 20% faster even in more complex tasks, a jump in training data and the introduction of two functions that bring it closer to human touch and memory.

Figure 2 is learning to generalize behaviors

A visual short-term memory that allows it to remember partially viewed barcodes and a force feedback system that works like a robotic sixth sense, adjusting grip strength based on the texture of the material, this means the Figure 2 squeezes on thin ones carefully, turns over deformable bags without tearing them, even taps on packages with bubble wrap so that the crumpled label can be scanned more accurately.

These adjustments raised the success rate in reading barcodes from 70% to 95% and all this increasing productivity instead of decreasing it, more than fulfilling a task Figue is learning to generalize behaviors, in a recent test some visual examples were enough for the robot to understand that an extended hand meant handing over the package and not placing it on the conveyor belt, this without any reprogramming, for now this is a demonstration but the evolution curve is clear if today it can endure an hour of work tomorrow it could be a shift complete and then a complete fleet replacing repetitive and tedious human tasks.