NDST: Neural Driving Style Transfer for Human-Like Vision-Based Autonomous Driving

TL;DR

The paper tackles the challenge of aligning autonomous driving with individual user comfort without compromising safety. It introduces Neural Driving Style Transfer (NDST), an NST-inspired framework that couples a Baseline Driving Model (BDM) with a Personalized Block (PB) to learn and apply a driver's unique acceleration/deceleration patterns. Training proceeds by first establishing a generic BDM, then freezing it while training a driver-specific PB using collected driving data, enabling per-user style transfer. In OSCAR-based simulations, NDST successfully reproduces distinct driving styles (A and B), improving perceived personalization while maintaining safety, suggesting a viable path to broader acceptance of autonomous driving systems.

Abstract

Autonomous Vehicles (AV) and Advanced Driver Assistant Systems (ADAS) prioritize safety over comfort. The intertwining factors of safety and comfort emerge as pivotal elements in ensuring the effectiveness of Autonomous Driving (AD). Users often experience discomfort when AV or ADAS drive the vehicle on their behalf. Providing a personalized human-like AD experience, tailored to match users' unique driving styles while adhering to safety prerequisites, presents a significant opportunity to boost the acceptance of AVs. This paper proposes a novel approach, Neural Driving Style Transfer (NDST), inspired by Neural Style Transfer (NST), to address this issue. NDST integrates a Personalized Block (PB) into the conventional Baseline Driving Model (BDM), allowing for the transfer of a user's unique driving style while adhering to safety parameters. The PB serves as a self-configuring system, learning and adapting to an individual's driving behavior without requiring modifications to the BDM. This approach enables the personalization of AV models, aligning the driving style more closely with user preferences while ensuring baseline safety critical actuation. Two contrasting driving styles (Style A and Style B) were used to validate the proposed NDST methodology, demonstrating its efficacy in transferring personal driving styles to the AV system. Our work highlights the potential of NDST to enhance user comfort in AVs by providing a personalized and familiar driving experience. The findings affirm the feasibility of integrating NDST into existing AV frameworks to bridge the gap between safety and individualized driving styles, promoting wider acceptance and improved user experiences.

Figures (11)

• Figure 1: (a) Neural Image Style Transfer (NST) gatys2015neural. NST extracts a style from one image and applies the style to the content of another image. (b) Neural Driving Style Transfer (NDST). NDST can produce personalized driving results by applying a user's driving style to an existing AD network.
• Figure 2: Structure of NDST Network. The NDST architecture comprises two main components: the Baseline Driving Model (BDM), a standard autonomous vehicle driving model based on NVIDIA's PilotNet, and the Personalized Block (PB), which customizes driving outputs according to individual driver styles. The BDM, depicted in blue, integrates image and speed inputs to generate standard vehicle actuations like steering, throttle, and braking through its five convolutional and three fully-connected layers. The PB, shown in red, inputs the BDM's predicted actions, road features, current speed, and the difference between target and current speeds, outputting personalized steering, throttle, and braking values.
• Figure 3: Neural driving style transfer (NDST) training process for AVs Personalization. The NDST training process begins with establishing a Baseline Driving Model (BDM). Subsequently, the Personalized Block (PB) is trained using data collected for each driver and the BDM, which is set to be non-trainable, ensuring that the PB adjusts and reflects the unique driving style of each driver.
• Figure 4: The training track used to collect training data from two different drivers.
• Figure 5: The test track used to test Neural Driving Style Transfer (NDST) system.