In this paper, a driving style recognition method based on self-supervised contrastive learning was developed. Traditional machine learning models cannot directly accept time series data of variables as inputs, and therefore, artificially constructed statistical variables are required. The driving style recognition model proposed in this paper can directly input the raw time-series data, which can more fully preserve the driving information. On the basis of three driving styles recognition, a style factor model was designed to make the driving style continuous, and the forgetting factor was introduced to synthesize the style factor under the most recent data and the style factor under the historical data. The implementation of personalized AEB strategies involved three steps: First, using the emergency factor as an indicator, the aggressive braking strategy line was obtained statistically. Then, according to the data from calm drivers and moderate drivers, the corresponding encoder-decoder models were established to predict the longitudinal relative distance and longitudinal relative velocity. Finally, based on the aggressive braking strategy line and the encoder-decoder models, the calm braking strategy and the moderate braking strategy were designed. The proposed AEB strategies achieved a higher rate of collision avoidance than the classic Mazda, Honda, and Berkeley strategies. When the risk of collision is eliminated from test samples, on average, compared with the aggressive strategy, the calm strategy is 1.05 s earlier, while the moderate strategy is 0.58 s earlier. Such results are more in line with the expectations of different drivers, and the personalized AEB system can improve driver acceptance.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Research on Personalized AEB Strategies Based on Self-Supervised Contrastive Learning


    Beteiligte:
    Li, Haotian (Autor:in) / Jin, Hui (Autor:in)


    Erscheinungsdatum :

    2024-02-01


    Format / Umfang :

    6656459 byte




    Medientyp :

    Aufsatz (Zeitschrift)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch







    Gated-ReRAM Based Strategies for On-Chip Supervised Learning

    Rush, Andrew J. / Jones, Alexander / Herrmann, Eric et al. | IEEE | 2019