CAR-Bench: AI Voice Assistant Evaluation Tasks for Automotive Environments

A benchmark dataset for evaluating AI voice assistants in realistic automotive contexts. It contains task configurations defining user personas, instructions, and vehicle context to test an agent's ability to use control tools, handle disambiguation, and avoid hallucinations. The dataset was created by johanneskirmayr and last updated on Hugging Face in February 2026.

Use Cases

• Benchmarking tool-call accuracy based on defined ground-truth sequences of tool calls.
• Evaluating disambiguation handling based on user personas and instructions.
• Testing for hallucination avoidance based on the initial vehicle and environment context.

Strengths

• Dataset is structured into task configs and mock data configs, suggesting an organized evaluation framework.
• Tasks include defined user personas, instructions, and vehicle context, providing a multi-faceted test scenario.

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

huggingface

Freshness

Last updated 2026-02-12 13:29:51; freshness should be verified.