TechniaHQRobot logoTechniaHQRobot
Reliability

Task Success Improvement S-Curve

The two comparable Figure study points are kept separate from the AGIBOT factory aggregate. No fitted learning curve is displayed.

Source observations through 2026-07-31. Analysis revised separately.Open all humanoid data charts

Data summary

What is included

The two comparable Figure study points are kept separate from the AGIBOT factory aggregate. No fitted learning curve is displayed.

Visible rows

3

Dataset scope

TechniaHQRobot source-linked dataset

Interactive infographic

Task Success Improvement S-Curve

The two comparable Figure study points are kept separate from the AGIBOT factory aggregate. No fitted learning curve is displayed.

  1. 10

    Software version
    Helix logistics scaling study
    Task
    Orient packages for barcode scanning on a moving conveyor
    Training data
    10 hours
    Reported success
    88.2 %
    Trials
    Not recorded
    Successes
    Not recorded
    Interventions
    Not recorded

    Exact trial count and confidence interval are not published in the cited article.

    Read source ↗

  2. 60

    Software version
    Helix logistics scaling study
    Task
    Orient packages for barcode scanning on a moving conveyor
    Training data
    60 hours
    Reported success
    94.4 %
    Trials
    Not recorded
    Successes
    Not recorded
    Interventions
    Not recorded

    Comparable with the 10-hour point in the same official study; exact trial count and confidence interval are not public.

    Read source ↗

  3. AgiBot A2 Ultra fleet

    Software version
    Not disclosed
    Task
    Multiple factory workflows
    Training data
    Not recorded
    Reported success
    99.99 %
    Trials
    64,828
    Successes
    Not recorded
    Interventions
    Not recorded

    The rounded success rate cannot be converted into exact success and failure counts.

    Read source ↗

Accessible data table

The table contains the same source-linked records used by the chart. It is the reference when labels overlap or a visual scale compresses values.

Source-linked data used in the infographic
RobotTaskSoftware versionTest dateTrialsSuccessesFailuresInterventionsEvidenceSource
Figure 02Orient packages for barcode scanning on a moving conveyorHelix logistics scaling study2025-06-07Not publicly disclosedNot publicly disclosedNot publicly disclosedNot publicly disclosedOfficial announcementFigure official Helix logistics scaling studyChecked 2026-07-31
Figure 02Orient packages for barcode scanning on a moving conveyorHelix logistics scaling study2025-06-07Not publicly disclosedNot publicly disclosedNot publicly disclosedNot publicly disclosedOfficial announcementFigure official Helix logistics scaling studyChecked 2026-07-31
AgiBot A2 Ultra fleetMultiple factory workflowsNot disclosed2026-06-2964828Not publicly disclosedNot publicly disclosedNot publicly disclosedCustomer confirmedAgiBot official Longcheer factory-operation reportChecked 2026-07-31

Methodology

How the dataset is built

Only points from a compatible protocol may share a curve. The two Figure points belong to one controlled study; the AgiBot factory aggregate is displayed separately. Rounded rates are not reverse-engineered into exact counts.

Editorial analysis

Two endpoints show a reported change

The Figure logistics records report success rates of 88.2% and 94.4% at 10 and 60 training hours. Within that study, the difference is 6.2 percentage points. The records do not provide a complete sequence of intermediate results.

The trial counts are missing

Without the number of trials and individual outcomes, the table cannot calculate a defensible uncertainty interval. Two endpoints also cannot establish the shape of a learning curve or where additional training stops producing useful gains.

The factory aggregate answers another question

AGIBOT’s factory row covers multiple workflows under a different protocol. It should not be inserted into the Figure series. Its rounded percentage cannot be reversed into exact success and failure counts, even where an aggregate task total is given.

Frequently asked questions

Can the page predict success after more training?

No. Two comparable endpoints and a separate factory aggregate do not establish a predictive learning curve.

Sources

Related humanoid infographics

Humanoid Robot Failure Pareto ChartA 24-Hour Humanoid Robot Operating CycleIntervention-Free Reliability CurveHumanoid Robot Companies by Country

Continue with the source-linked humanoid catalog

Open individual robot records for manufacturer, origin, dimensions, availability, public price status and official product links.

Browse humanoid robotsRead the main guide