HEADD is a dataset of natural language explanations elicited from online participants via Prolific with corresponding annotations for each explanation similarly given by online (but different) participants. The data is about driving scenarios in which the behaviour and driving decisions of a single blue agent need to be explained, while the scenarios contain various other agents and environmental elements that influence the behaviour of the blue agent. The dataset contains 14 unique scenarios with qualitatively distinct and interesting driving scenarios including simulated video recordings, ASAM OpenDrive maps, and ASAM OpenScenario descriptions. In addition, HEADD includes 1308 explanations in natural language with 4 explanatory modes (descriptive, teleological, mechanistic, counterfactual) from 54 participants in each of the 14 scenarios, of which 947 non-descriptive explanations are annotated with 5 unique annotations regarding the causal content and trustworthiness of the explanations under the various circumstances in the scenarios.
Date made available26 Jan 2024
PublisherEdinburgh DataShare
Geographical coverageUS,UNITED STATES

Cite this