kantine2026apache-2.0

BotFails

A multimodal dataset for robotic failure detection in manipulation tasks, containing 144 annotated anomalous episodes across 10 domestic and industrial tasks with vision, proprioception, and language annotations.

Downloads0
Episodes144 annotated anomalous episodes plus nominal demonstrations

Why This Matters for Physical AI

BotFails enables research on robotic failure detection and anomaly localization, which is critical for building safe and reliable autonomous robotic systems that can identify and respond to execution errors in real-world manipulation tasks.

Technical Profile

Modalities
rgbproprioceptionlanguage
Robot Embodiments
LeRobot
Environment
lab
Task Types
manipulationgraspingpick_and_placepouringsortingassembly
Episodes
144 annotated anomalous episodes plus nominal demonstrations
Data Format
LeRobot
Annotation Types
language_instructionsanomaly_labelsfailure_type_labels
License
apache-2.0
Part of the BotFails family

Access

Need custom rgb data?

Claru builds purpose-built datasets for lab applications with dense human annotations and quality assurance.

Request a Sample Pack

Related Datasets