Figures AI, one of the cutting companies in the development of humanoid robots, recently showed an important jump forward for his Figure 02 model, which shows what the machine is now able to sort packages at a speed and agility that approach the human.
The secret of this reinforcement lies in the robot visa man of the Visa control system, Helix, which received crucial updates. The engineers integrated a visual memory module in the short term and a Force Feedback system. With this combination the robot can have A more advanced perception of the area: Memory enables him to remember partial glimpses of a barcode that was seen a moment earlier, so that the optimum rotation is planned to expose it to the scanner. The Force Feedback, on the other hand, acts as a kind of feeling of touch, and leads the grip to be more delicate or decisive according to the need and to enable a quick and precise release.
Thanks to these refinements, the performance has changed radically. In a new video, not edited and lasts an hour, you can see a single figure 02 at work without interruption, able to process each package in approximately 4.05 seconds. Look at yourself how it moves and with what precision.
https://www.youtube.com/watch?v=LKC2Y0YB89U
This is an improvement of 20% compared to the performance a few months ago, an even more remarkable result if we consider the increase in the complexity of the tasks. The success rate in the scan of barcodes went from a modest 70% to an impressive 95%. This huge step forward was also possible thanks to a huge increase in training data, which grew six times, from ten to sixty hours of demonstrations.
These figures translate into a very concrete ability to handle objects that were problematic for the car until recently. In addition to the common cardboard boxes, the robot now manages with ease, even no -rigid packages such as deformable plastic bags and flat padded envelopes. With the helix system, figures 02 can adjust his strategy in real time: It can carefully squeeze the thin envelopes, give a capetto to the soft bags to turn them upside down or to flatten a plastic from a package to make a perfectly readable label. Perhaps the most interesting aspect is the ability to generalize.
With only a handful of additional examples, the robot has learned to recognize the tense hand of a human operator and to deliver a package without a specific programming for this task. Although the mass distribution is not yet at the gates, the process is clear. Figures, together with competitors such as Optimus by Tesla and figure of agility, show that constant increase in data and in the design of neural networks fill the gap between a prototype and a tireless employee.
#Humanoid #Robot #Figures #Sort #packages #person #seconds #Video