EgoTouch: Egocentric Hand-Object Interaction with Tactile Sensing

EgoTouch is a large-scale multi-view tactile dataset for egocentric hand-object interaction, converted to HDF5 format. It contains 208 diverse manipulation tasks across 1,891 episodes recorded in indoor and outdoor environments. The dataset provides synchronized egocentric and dual wrist camera videos, bimanual hand pose annotations, and dense continuous pressure maps from wearable tactile sensors.

Use Cases

• Train robotic manipulation models based on synchronized video and tactile sensor data.
• Develop hand pose estimation algorithms using the provided bimanual hand pose annotations.
• Research multimodal fusion techniques for egocentric vision and continuous pressure maps.
• Benchmark models for understanding human-object interaction in diverse indoor and outdoor environments.

Strengths

• Contains 1,891 recorded episodes, providing a substantial number of interaction sequences.
• Covers 208 diverse manipulation tasks, suggesting variety in the types of interactions captured.
• Includes synchronized data streams from multiple modalities: egocentric video, dual wrist cameras, hand pose, and tactile pressure.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

TouchAnything project, converted and hosted by user 'zhenyuxie-zhzh' on Hugging Face.

Collection Method

Likely collected through recordings of human subjects performing manipulation tasks with wearable sensors and cameras.

Freshness

Last updated 2026-05-28 02:14:03; freshness should be verified.