Adaptive driver modelling in ADAS to improve user acceptance: A study using naturalistic data

Abstract

Accurate understanding of driver behaviour is crucial for future Advanced Driver Assistance Systems (ADAS) and autonomous driving. For user acceptance it is important that ADAS respect individual driving styles and adapt accordingly. Using data collected during a naturalistic driving study carried out at the University of Southampton, we assess existing models of driver acceleration and speed choice during car following and when cornering. We observe that existing models of driver behaviour that specify a preferred inter-vehicle spacing in car-following situations appear to be too prescriptive, with a wide range of acceptable spacings visible in the naturalistic data. Bounds on lateral acceleration during cornering from the literature are visible in the data, but appear to be influenced by the minimum cornering radii specified in design codes for UK roadway geometry. This analysis of existing driver models is used to suggest a small set of parameters that are sufficient to characterise driver behaviour in car-following and curve driving, which may be estimated in real-time by an ADAS to adapt to changing driver behaviour. Finally, we discuss applications to adaptive ADAS with the objectives of improving road safety and promoting eco-driving, and suggest directions for future research.

An FPGA-Based Neuro-Fuzzy Sensor for Personalized Driving Assistance

“…First, the selected features THWRMS, TETH, and TITH features were computed for each of the 176 driving segments and normalized into the [0,1] range. According to the minimum following safety threshold of [16] and the selected THW in the work by the authors of [19], TETH and TITH were calculated for a critical value of THW*=1.5 s. After that, three-group k-means clustering of those segments was performed. The obtained cluster structure is depicted in Figure 5.…”

“…As can be seen in Figure 13, for each of the models, the predicted trueTHW^i was directly proportional with THW¯RMSi and inversely proportional with TITH. Note that, for the Cluster 1 model, the plane was saturated to THW^i=1 s to assure that the personalized THW value never took a value lower than the minimal safe THW values [16].…”

“…The conjunction of all of these intelligent sensors, together with high-speed wireless communications, have allowed car-makers for the first time to develop intelligent vehicles with automated driving systems [15]. The acceptance of ADAS by drivers mainly depends on the engineering factors of the system, predefined by technicians and implied by system functional specifications [16]. Thus, for the longitudinal control of vehicles, the time gap with the precedent vehicle, known as the time headway (THW), is monitored by means of radar sensors, considering the longitudinal area of safe travel in relation to the predefined stop distance parameter limits set by the manufacturer.…”

Fitting Cornering Speed Models with One-Class Support Vector Machines

“…Time-series data from GPS was analysed by extracting all local curvature maxima and the corresponding observed velocity, giving a total of 7384 cornering events that we denote by a pair (v i , a i ) of cornering speed and lateral acceleration values. For more detailed information on the collection and processing of this cornering data, we refer readers to [18], which describes the data processing and analysis, and [19], which describes the data collection device used.…”

Incorporating Driver Preferences Into Eco-Driving Assistance Systems Using Optimal Control

“…User acceptance is a well-known issue with existing automotive ADAS, for example for collision-warning systems [19] that rely on knowledge of typical vehicle-following behaviour. Based on the large variations between drivers observed in naturalistic studies, one suggestion to improve user acceptance of ADAS is to make the system adaptive, adjusting to the driver by estimating parameters representing their driving style in real-time [20]. For eco-driving assistance systems this possibility has already been explored for the specific scenario of approaching intersections [21].…”