Adaptive cruise control has been an important function in modern vehicles, and has proven to be helpful for assisted driving. The main challenges involve accurate gap prediction between the ego and preceding vehicles, as well as personalizing the driving behaviour for different kinds of drivers and/or cars. Correspondingly, in this paper, we make the following contributions: (1) we propose GAPFoRMER which combines the Transformer and RNN architectures to better model and personalize driving behaviour; (2) make necessary modifications to the Transformer attention mechanism for scaling to long driving contexts in a resource-efficient manner; and (3) propose an architecture-agnostic model training regime, Horizon which improves generalization by incorporating a time-horizon and makes the models more accurate and robust. Detailed experiments on both public and proprietary datasets demonstrate that GAPFoRMER can be up to 50% more accurate when compared to other ACC baselines, demonstrating its efficacy and potential for real-world application.
GapFormer: Fast Autoregressive Transformers meet RNNs for Personalized Adaptive Cruise Control
08.10.2022
1252274 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Method and system for personalized car following with transformers and RNNs
Europäisches Patentamt | 2025
|METHOD AND SYSTEM FOR PERSONALIZED CAR FOLLOWING WITH TRANSFORMERS AND RNNS
Europäisches Patentamt | 2024
|Personalized Adaptive Cruise Control Considering Drivers' Characteristics
British Library Conference Proceedings | 2018
|