Driver's trust assessment based on situational awareness under human-machine collaboration driving

Abstract

The human-machine co-driving can effectively take into account the superior characteristics of human driver and agent, and generate the ‘1 + 1>2’ collaboration mode. However, drivers with autonomous consciousness and different behaviors actively intervene in the control of vehicles, which will trigger the conflict between human and intelligent driving system. The human-machine mutual trust has gradually developed one of the key technologies to mitigate human-machine conflicts. In this study, from the perspective of machine trust towards human, a comprehensive trust evaluation model (CTEM) for intelligent systems towards driver intervention behavior was established. Based on the driver's situational awareness (SA) recovery process, the model focused on the hierarchical decision-making model from perception to cognition. A residual convolutional neural network based on attention mechanism was proposed to identify active intervention and accidental touches. Then the perceived trust assessment (PTA) model was established on the basis of visual geometry group16 (VGG16) network. For the cognitive trust assessment (CTA) model, the long short-term memory (LSTM) codec structure was employed to predict the vehicle trajectory, and the risk field was structured to quantify the risk value of the future trajectory. The CTEM model was constructed by integrating the PTA and CTA models, which serve as its foundational components. Finally, the driving simulator experiments was implement to verify the proposed model, and the results demonstrated that the CTEM could availably distinguish and assess the driver's SA ability. The construction of trust evaluation model will provide effective support for the improvement of human-machine co-driving safety.