AutoTherm: A Dataset and Benchmark for Thermal Comfort Estimation Indoors and in Vehicles

TL;DR

AutoTherm introduces two temporal, multimodal datasets for indoor and in-vehicle thermal comfort, comprising 31 input signals and self-labeled judgments from participants. The authors demonstrate that time-series data processed with recurrent models outperform single-vector baselines, and provide extensive validations against public datasets to quantify generalization gaps. They reveal a notable divergence between indoor and automotive thermal comfort estimation, and offer a practical data-collection blueprint along with feature analyses to guide future state-recognition work for adaptive HVAC and vehicle interiors. The publicly released dataset and code aim to enable digital twin-inspired interaction design and real-time climate control in intelligent environments.

Abstract

Thermal comfort inside buildings is a well-studied field where human judgment for thermal comfort is collected and may be used for automatic thermal comfort estimation. However, indoor scenarios are rather static in terms of thermal state changes and, thus, cannot be applied to dynamic conditions, e.g., inside a vehicle. In this work, we present our findings of a gap between building and in-vehicle scenarios regarding thermal comfort estimation. We provide evidence by comparing deep neural classifiers for thermal comfort estimation for indoor and in-vehicle conditions. Further, we introduce a temporal dataset for indoor predictions incorporating 31 input signals and self-labeled user ratings by 18 subjects in a self-built climatic chamber. For in-vehicle scenarios, we acquired a second dataset featuring human judgments from 20 subjects in a BMW 3 Series. Our experimental results indicate superior performance for estimations from time series data over single vector input. Leveraging modern machine learning architectures enables us to recognize human thermal comfort states and estimate future states automatically. We provide details on training a recurrent network-based classifier and perform an initial performance benchmark of the proposed dataset. Ultimately, we compare our collected dataset to publicly available thermal comfort datasets.

Figures (14)

• Figure 1: Overview of the climate chamber.
• Figure 2: Left: Labels given during heating. Right: Labels for the cooling phase. In line with previous research (e.g., ciuhaEffectThermalTransience2019), thermal comfort is established at a lower ambient temperature during heating than during cooling.
• Figure 3: Sensory setup inside the BMW $3$ F31. The Arduino board was mounted on top of the middle A/C vents. The displays were covered to hide temperature information from participants. In-vehicle setup did not include a Kinect camera.
• Figure 4: Labels given during in-vehicle study.
• Figure 5: Feature importance ranking based on mean accuracy impurity decrease of an RF classifier. According to this metric, the five most important features were ambient temperature, relative humidity, radiation temperature, skin temperature, and galvanic skin response, marked in green.