SWITCH-Basic V1: Tangible Interface Interaction Data for Embodied Agents

SWITCH-Basic V1 Open contains between 1,000 and 10,000 records of real-world Tangible Computer Interface (TCI) interaction data for embodied agents. Developed by BAAI-Agents and released in early 2026, this multimodal collection includes images and videos of physical interface interactions and verification tasks.

Use Cases

• Training embodied agents to interact with Tangible Computer Interfaces (TCI) using image and video inputs
• Benchmarking visual verification capabilities for robotic manipulation of physical controls
• Developing Visual Question Answering (VQA) models to identify physical interface states

Strengths

• Contains 1,000 to 10,000 multimodal records
• Focuses on real-world Tangible Computer Interfaces (TCI) rather than simulated environments
• Includes both image and video modalities for temporal interaction analysis
• Supports Visual Question Answering (VQA) tasks in a robotics context

Limitations

• Limited to a 30% subset of the full benchmark data
• Restricted to non-commercial use under the CC BY-NC 4.0 license
• Domain-specific focus on tangible interfaces may not generalize to digital-only GUI agents

Provenance

Source

BAAI-Agents (Arxiv 2511.17649)

Collection Method

Systematic collection and annotation of real-world TCI interaction data

Freshness

Last updated February 2026.