MIT Advanced Vehicle Technology Study: Large-Scale Naturalistic Driving Study of Driver Behavior and Interaction with Automation (1711.06976v4)

Abstract: For the foreseeble future, human beings will likely remain an integral part of the driving task, monitoring the AI system as it performs anywhere from just over 0% to just under 100% of the driving. The governing objectives of the MIT Autonomous Vehicle Technology (MIT-AVT) study are to (1) undertake large-scale real-world driving data collection that includes high-definition video to fuel the development of deep learning based internal and external perception systems, (2) gain a holistic understanding of how human beings interact with vehicle automation technology by integrating video data with vehicle state data, driver characteristics, mental models, and self-reported experiences with technology, and (3) identify how technology and other factors related to automation adoption and use can be improved in ways that save lives. In pursuing these objectives, we have instrumented 23 Tesla Model S and Model X vehicles, 2 Volvo S90 vehicles, 2 Range Rover Evoque, and 2 Cadillac CT6 vehicles for both long-term (over a year per driver) and medium term (one month per driver) naturalistic driving data collection. Furthermore, we are continually developing new methods for analysis of the massive-scale dataset collected from the instrumented vehicle fleet. The recorded data streams include IMU, GPS, CAN messages, and high-definition video streams of the driver face, the driver cabin, the forward roadway, and the instrument cluster (on select vehicles). The study is on-going and growing. To date, we have 122 participants, 15,610 days of participation, 511,638 miles, and 7.1 billion video frames. This paper presents the design of the study, the data collection hardware, the processing of the data, and the computer vision algorithms currently being used to extract actionable knowledge from the data.

• The paper presents comprehensive naturalistic driving data that uncover critical insights into human and automation interactions.
• Robust multi-sensor systems and advanced machine learning techniques underpin the study’s methodology to synchronize diverse data streams.
• Findings from the study guide improvements in safety features, user interfaces, and regulatory frameworks for advancing autonomous vehicle technology.

Insights on the MIT Advanced Vehicle Technology Study

The paper "MIT Advanced Vehicle Technology Study" presents a comprehensive overview of a large-scale naturalistic driving paper focused on understanding driver behavior and interaction with automation technologies. This paper, hereafter referred to as MIT-AVT, aims to address the substantial gap between the current state of autonomous vehicle technology and fully autonomous driving.

Background and Objectives

This endeavor aims to bridge the technical complexities inherent in achieving full vehicle autonomy by integrating human elements into the equation. Historically, events like the DARPA Urban Challenge have portrayed autonomous driving as a nearly attainable goal. Yet, real-world driving introduces a multitude of variables including human unpredictability, environmental conditions, and technology limitations. The MIT-AVT paper aims to tackle these challenges by continuously collecting and analyzing extensive naturalistic driving data, thus prioritizing the human-automation interaction over theoretical perfection in technology.

The design of MIT-AVT harnesses a wide spectrum of vehicles and technologies, encompassing 23 Tesla Model S and Model X vehicles and other models from Volvo, Range Rover, and Cadillac. The paper endeavors to collect long-term (over a year) and medium-term (one month) driving data capturing diverse environmental settings, driver behaviors, and technological interactions.

Methodology

Key methodological pillars underpinning the paper include:

1. Data Acquisition: The instrumentation of the vehicles involves capturing data from multiple sources, including IMU, GPS, CAN messages, and high-definition camera feeds. These sources provide a holistic view of the driver, vehicle, and environment interactions.
2. Software and Hardware Infrastructure: A robust data recording system known as RIDER has been developed to process and synchronize data across sensors efficiently. This system ensures the seamless integration of data streams for subsequent analysis.
3. Data Analysis: A high focus is laid on the application of computer vision and machine learning to derive meaningful insights. This involves using deep learning for image classification, object detection, and semantic segmentation from the collected data.
4. Real-world Data Utilization: The paper uniquely emphasizes the "long-tail" of driving data. Instead of focusing solely on crash or near-crash events, it analyzes more mundane and frequent interactions between drivers and autonomy-supportive systems.

Implications and Future Work

The MIT-AVT paper represents a significant step in understanding how drivers interact with semi-autonomous systems. The comprehensive data collection enables a nuanced analysis of how automation influences driving behavior, which can inform the design of future driving assistance technologies. Moreover, the consortium model, involving collaboration between academia, automotive manufacturers, and other stakeholders, signifies an important pathway for shared learning and innovation in the field.

Practically, insights from the paper could lead to improvements in safety features, user interfaces, and trust in automation, fostering a smoother transition to fully autonomous vehicles. Assessing broader social impacts, such as societal acceptance and regulatory developments, will also be crucial.

Looking forward, extensions of this work could include developing predictive models for driver behavior under various degrees of automation or expanding the scope to include a wider variety of vehicle types and driving environments. It may also stimulate further interdisciplinary research efforts combining robotics, human factors, and computer science to accelerate advancements in vehicle automation technologies.

In summary, the MIT-AVT paper stands as a substantial contribution to the ongoing efforts in the autonomous driving space, delivering valuable empirical insights into the complex interplay between humans and automated vehicle systems.