OBSERVED FAILURE ≠ CONFIRMED CAUSE

Physical AI Failure Database

Review documented Physical AI task failures and public failure taxonomies without guessing causes or turning isolated events into universal reliability claims.

7 retained recordsLast verified: August 13, 2026Primary-source fields preserved per record

Failure typeStatus

7 of 7 records shownFilters are local and do not create crawlable URLs

TurboVLA

Research result

Date: 2026-07-29
Model: TurboVLA
Robot: AgileX Piper
Task: Grab the roller
Environment: Controlled lab tabletop
Failure type: Task failure; specific cause not disclosed in aggregate result
Observed failure: 3 of 40 evaluation trials were unsuccessful, derived exactly from the published 92.5% success rate and 40-trial denominator.
Recovery: Not reported
Human intervention: Not reported
Company explanation: No single cause assigned to the aggregate failures
Confirmed cause: Unknown
Status: Research evaluation
Source: TurboVLA project

TurboVLA

Research result

Date: 2026-07-29
Model: TurboVLA
Robot: AgileX Piper
Task: Move the playing card
Environment: Controlled lab tabletop
Failure type: Task failure; specific cause not disclosed in aggregate result
Observed failure: 8 of 40 evaluation trials were unsuccessful, derived exactly from the published 80% success rate and 40-trial denominator.
Recovery: Not reported
Human intervention: Not reported
Company explanation: No single cause assigned to the aggregate failures
Confirmed cause: Unknown
Status: Research evaluation
Source: TurboVLA project

TurboVLA

Research result

Date: 2026-07-29
Model: TurboVLA
Robot: AgileX Piper
Task: Press the stapler
Environment: Controlled lab tabletop
Failure type: Task failure; specific cause not disclosed in aggregate result
Observed failure: 4 of 40 evaluation trials were unsuccessful, derived exactly from the published 90% success rate and 40-trial denominator.
Recovery: Not reported
Human intervention: Not reported
Company explanation: No single cause assigned to the aggregate failures
Confirmed cause: Unknown
Status: Research evaluation
Source: TurboVLA project

TurboVLA

Research result

Date: 2026-07-29
Model: TurboVLA
Robot: AgileX Piper
Task: Stack three bowls
Environment: Controlled lab tabletop
Failure type: Task failure; specific cause not disclosed in aggregate result
Observed failure: 5 of 40 evaluation trials were unsuccessful, derived exactly from the published 87.5% success rate and 40-trial denominator.
Recovery: Not reported
Human intervention: Not reported
Company explanation: No single cause assigned to the aggregate failures
Confirmed cause: Unknown
Status: Research evaluation
Source: TurboVLA project

Not disclosed as one model in the retained taxonomy

Official announcement

Date: 2026
Model: Not disclosed as one model in the retained taxonomy
Robot: AgiBot World dataset
Task: Manipulation trajectories
Environment: Public dataset taxonomy
Failure type: Grasp failure
Observed failure: AgiBot publicly names grasp failure as an error-trajectory category; comparable frequency is not published in the retained source.
Recovery: Not disclosed
Human intervention: Dataset/error-trajectory record
Company explanation: Official dataset taxonomy
Confirmed cause: Not established
Status: Official taxonomy
Source: AgiBot World 2026 Theme 2

Not disclosed as one model in the retained taxonomy

Official announcement

Date: 2026
Model: Not disclosed as one model in the retained taxonomy
Robot: AgiBot World dataset
Task: Manipulation trajectories
Environment: Public dataset taxonomy
Failure type: Collision
Observed failure: Collision is a named public error-trajectory category. The cited source does not publish a comparable frequency.
Recovery: Not disclosed
Human intervention: Dataset/error-trajectory record
Company explanation: Official dataset taxonomy
Confirmed cause: Not established
Status: Official taxonomy
Source: AgiBot World 2026 Theme 2

Not disclosed as one model in the retained taxonomy

Official announcement

Date: 2026
Model: Not disclosed as one model in the retained taxonomy
Robot: AgiBot World dataset
Task: Manipulation trajectories
Environment: Public dataset taxonomy
Failure type: Object drop
Observed failure: Object drop is a named public error-trajectory category. The cited source does not publish a comparable frequency.
Recovery: Not disclosed
Human intervention: Dataset/error-trajectory record
Company explanation: Official dataset taxonomy
Confirmed cause: Not established
Status: Official taxonomy
Source: AgiBot World 2026 Theme 2

Date	Model	Robot	Task	Environment	Failure type	Observed failure	Recovery	Human intervention	Company explanation	Confirmed cause	Status	Source
2026-07-29	TurboVLA	AgileX Piper	Grab the roller	Controlled lab tabletop	Task failure; specific cause not disclosed in aggregate result	3 of 40 evaluation trials were unsuccessful, derived exactly from the published 92.5% success rate and 40-trial denominator.	Not reported	Not reported	No single cause assigned to the aggregate failures	Unknown	Research evaluation	TurboVLA project
2026-07-29	TurboVLA	AgileX Piper	Move the playing card	Controlled lab tabletop	Task failure; specific cause not disclosed in aggregate result	8 of 40 evaluation trials were unsuccessful, derived exactly from the published 80% success rate and 40-trial denominator.	Not reported	Not reported	No single cause assigned to the aggregate failures	Unknown	Research evaluation	TurboVLA project
2026-07-29	TurboVLA	AgileX Piper	Press the stapler	Controlled lab tabletop	Task failure; specific cause not disclosed in aggregate result	4 of 40 evaluation trials were unsuccessful, derived exactly from the published 90% success rate and 40-trial denominator.	Not reported	Not reported	No single cause assigned to the aggregate failures	Unknown	Research evaluation	TurboVLA project
2026-07-29	TurboVLA	AgileX Piper	Stack three bowls	Controlled lab tabletop	Task failure; specific cause not disclosed in aggregate result	5 of 40 evaluation trials were unsuccessful, derived exactly from the published 87.5% success rate and 40-trial denominator.	Not reported	Not reported	No single cause assigned to the aggregate failures	Unknown	Research evaluation	TurboVLA project
2026	Not disclosed as one model in the retained taxonomy	AgiBot World dataset	Manipulation trajectories	Public dataset taxonomy	Grasp failure	AgiBot publicly names grasp failure as an error-trajectory category; comparable frequency is not published in the retained source.	Not disclosed	Dataset/error-trajectory record	Official dataset taxonomy	Not established	Official taxonomy	AgiBot World 2026 Theme 2
2026	Not disclosed as one model in the retained taxonomy	AgiBot World dataset	Manipulation trajectories	Public dataset taxonomy	Collision	Collision is a named public error-trajectory category. The cited source does not publish a comparable frequency.	Not disclosed	Dataset/error-trajectory record	Official dataset taxonomy	Not established	Official taxonomy	AgiBot World 2026 Theme 2
2026	Not disclosed as one model in the retained taxonomy	AgiBot World dataset	Manipulation trajectories	Public dataset taxonomy	Object drop	Object drop is a named public error-trajectory category. The cited source does not publish a comparable frequency.	Not disclosed	Dataset/error-trajectory record	Official dataset taxonomy	Not established	Official taxonomy	AgiBot World 2026 Theme 2

Methodology

Observed failure, company explanation and independently confirmed cause remain separate fields. When a benchmark publishes only success rate and trial count, the number of unsuccessful trials can be calculated exactly, but the cause stays Unknown unless documented.

Included

Published model, robot, benchmark, deployment or failure evidence that can be tied to a retained source.

Not inferred

Unknown autonomy, latency, trial counts, deployment scale and failure causes remain unknown when the source does not disclose them.

Update policy

Versions and historical records remain distinguishable. A new release does not silently overwrite an older model or evaluation condition.