Skip navigation
X Square ran a fully autonomous logistics livestream: sorting parcels and flipping the barcode side…
0:00 / 0:46

X Square ran a fully autonomous logistics livestream: sorting parcels and flipping the barcode side…

The Humanoid Hub

@TheHumanoidHub116K followers

Visit on X
281 likesDownloadOriginal post

72,067 views

X Square ran a fully autonomous logistics livestream: sorting parcels and flipping the barcode side up for the scanner. - 1,816 parcels/hour (~2 seconds per parcel) - 98% accuracy - ~45% higher throughput than Figure's sub-3 seconds per parcel pace from its multiday livestream in May. It's driven by X Square's WALL-B embodied AI model, which continuously reacts to an unstructured pile of parcels, sorts and flips them, and recovers from failed grasps and parcel jams to keep the line moving.

14 Comments

Comments from the original post on X (Twitter)

  • PrismaX's profile picture

    @PrismaXai

    1,816 per hour is a real throughput number 🦾

    5 likes

  • Krafters Robotics's profile picture

    @krafters_xyz

    1,816 parcels/hour is useful. What made up the failures, and did any require a person to reset the line? That tells operators how much of the hour ran without help.

    2 likes

  • NexoraAI's profile picture

    @nexora_robot

    Would be interesting to see how it handles damaged boxes, weird shapes, and reflective packaging

    2 likes

  • Remcon Equipment's profile picture

    @LibertyForgeMfg

    Every conveyor guy watching this is looking at the table, not the robots.

    1 like

  • cetusian's profile picture

    @cetusian

    98 percent at 1,816 an hour is about 36 parcels going the wrong way every hour. the throughput number only means something next to what recovery costs. does a miss get caught downstream or does it just arrive somewhere wrong?

    1 like

  • Frontier Signal's profile picture

    @icpburn

    Useful, but not matched: two fixed arms vs a rotating three-humanoid fleet; 1 hour vs 200. Buyers need cost per successful parcel, changeover, exception recovery, floor space and sustained uptime. Near term, arms/grippers/vision may monetize before humanoid bodies.

    1 like

  • Ben Silone's profile picture

    @bsilone

    @aizorr0 A specialized machine isn’t really comparable to a humanoid robot

    1 like

  • ヒカル's profile picture

    @hikaru_Apex

    ついに2足ロボットは作業に向かないのに気付いたか

  • Alexander Dixon's profile picture

    @AlexanderDixon_

    It’s coming, start becoming interesting now

  • SEAR's profile picture

    @joinsear

    Fast sorting is nice

  • COCOBOY's profile picture

    @dlwndrbs1

    피규어보다 훨빠르노

  • Stephan Jaeckel's profile picture

    @StephanJaeckel

    Obviously much better performance by design, than a humanoid offers. One knows without seeing the data. But the process as such is still broken by design. Whether a human, a humanoid or two robotic arms are applied to it doesn't change that.

  • Mirrorworld AI's profile picture

    @mirrorworld_ai

    2 seconds per parcel with recovery logic. That's where sim-to-real transfer actually shows.

  • Benjamin Peng's profile picture

    @BinPeng22

    The recovery behavior may be more important than the raw speed. Real-world robotics gets valuable when it can handle messy inputs, failed grasps, and jams without constant human intervention.