SWITCH: Benchmark for Tangible Interface Actions in Egocentric Scenarios

SWITCH (Semantic World Interface Tasks for Control & Handling) is a multimodal embodied-interaction benchmark created by BAAI-Agents. It focuses on understanding, modeling, and evaluating actions over Tangible Control Interfaces (TCIs) like appliance panels and lighting controls in real-world, egocentric scenarios. The dataset was last updated on 2026-06-18.

Use Cases

• Benchmarking action recognition models based on interactions with tangible control interfaces.
• Training embodied AI agents for long-horizon tasks involving real-world devices.
• Evaluating multimodal understanding of egocentric scenarios that include everyday interfaces.

Strengths

• Focuses on a specific and practical domain: tangible control interfaces in real-world settings.
• Designed as a benchmark for embodied AI, providing a structured evaluation target.
• Created by BAAI-Agents, an organization associated with AI research.

Limitations

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

Provenance

Source

BAAI-Agents via Hugging Face

Collection Method

Likely collected from egocentric real-world scenarios involving tangible interfaces.

Freshness

Last updated 2026-06-18 07:19:48; freshness should be verified.