FAILURE EVIDENCE

Humanoid Robot Failure Database

Documented failure evidence, denominators and intervention notes without inferring fleet-wide reliability from one event.

Last verified: August 13, 2026• 7 retained records

Evidence rows

Failure modes

Rows with denominator

Required for rate-like interpretation

Robot	Failure / observation	Count	Exposure denominator	Intervention	Observation type	Evidence	Source
AgiBot A2 Ultra AgiBot	Fall	0	5,100 hours of walking validation	None reported for this failure mode	Manufacturer validation claim	official-specification	Open source
AgiBot A2 Ultra AgiBot	Operational anomaly	0	1,300 continuous walking hours	None reported	Manufacturer validation claim	official-specification	Open source
AgiBot World dataset AgiBot	Missed grasp	Not publicly disclosed	Not published	Error trajectory retained in dataset	Public dataset taxonomy	official-announcement	Open source
AgiBot World dataset AgiBot	Collision	Not publicly disclosed	Not published	Error trajectory retained in dataset	Public dataset taxonomy	official-announcement	Open source
AgiBot World dataset AgiBot	Object drop	Not publicly disclosed	Not published	Error trajectory retained in dataset	Public dataset taxonomy	official-announcement	Open source
AgiBot World dataset AgiBot	Unstable contact	Not publicly disclosed	Not published	Error trajectory retained in dataset	Public dataset taxonomy	official-announcement	Open source
AgiBot World dataset AgiBot	Liquid splash	Not publicly disclosed	Not published	Error trajectory retained in dataset	Public dataset taxonomy	official-announcement	Open source

A manufacturer validation claim such as “0 falls in 5,100 hours” is displayed as that specific claim. It is not converted into a fleet-wide reliability score.

METHODOLOGY

The retained dataset mixes validation claims and published failure taxonomies. The interface labels observation type and denominator so a zero-failure validation claim is not presented as field reliability evidence.