Accelerating the Future...
- Award Categories: Transportation
- Award Year: 2019
- Nominee URL: https://www.dreyev.com/
Despite substantial investments in highway safety and automotive technologies, crashes, injuries, and deaths on US roads are on the rise (NHTSA, 2018b), and in approximately 94 percent of crashes, the critical factor in the crash is a driver action or error (NHTSA, 2015). Driver distraction is a major factor in serious crashes, with 9 percent of fatal crashes involving a distracted driver (NHTSA, 2018), and young drivers aged 15 to 19 are 50 percent more likely to be involved in distraction-related crashes than drivers aged 20 and older. While many States have attempted to mitigate driver distraction through laws to limit the use of mobile phones, research shows that these bans have not reduced crashes (HLDI, 2009). In the past 3 years, auto claim severity in the United States has grown significantly. The most troubling trend was the restarted growth in fatalities. Since 1930s auto accident fatalities have been steadily declining year over year, but this trend reversed in 2014. Distracted driving is considered to be one of the major causes of fatal crashes. The increase in fatalities has been largely among bicyclists, motorcyclists, and pedestrians — all of whom are easier to miss from the driver’s seat, especially if a driver is glancing up from a phone rather than concentrating on the road. For instance, pedestrians’ fatalities increased by 22 %. As of 2017, the economic toll of distracted driving on the US reached $129B, which is surpassed only by drunk driving cost of $199B. The eventual proliferation of autonomous vehicles may solve the issue of both distracted and drunk driving, but it will take decades before we see widespread benefits. In addition to the human toll, crashes have a substantial impact on the economy and cause non-recurrent congestion negatively impacting transportation networks and straining transportation agency resources.
Dreyev is a digital co-pilot that evaluates drivers for distracted, drowsy or reckless behavior. It includes an in-vehicle dual camera, powered by computer vision and machine learning in a smartphone-based application. Dreyev deploys behavioral AI to analyze emerging driving risks, monitor driver attention and alertness of the driver through head pose tracking and generate effective alerts before a collision becomes inevitable. The dreyev platform analyzes driver conditions such as eyelid closure, head nodding and yawning to detect drowsy driving signs, in addition to distracted driving. Vehicle telemetry, in-cabin camera inputs and road scene analysis identify relevant inputs on unsafe driving (e.g. speeding), road hazards, and driver ID to enable a holistic view on the driving context, and determine the appropriate communications to the driver. Communications (warning, alerts) with the driver use acoustic and verbal clues. Occasional updates on driving performance are communicated to the driver at safe times, upon request.
Dreyev is the most advanced solution to prevent crashes caused by inattentive driving. Unlike existing solutions, dreyev creates driving behavior models, anticipates upcoming risks and generates personalized, real-time, urgency-triggered alerts. As a unique technology, it evaluates driver's reaction time and quality of corrective actions in response to warnings and alerts generated by the virtual copilot, in order to generate an accurate profile of a driver's attitude to risk and ability to cope with it. Uniqueness Dreyev’s approach is unique in the multidimensional assessment of the risk the driver is exposed to and their ability to manage it. Most importantly, alerts are generated in real time, personalized to maximize driver’s responsiveness, to allow the driver to correct his/her behavior and PREVENT accidents in the first place. Unlike the currently existing driver attention monitoring solutions, dreyev creates personalized driving behavior models, by evaluating responsiveness of the driver and effectiveness of the corrective actions, in order to maximize timeliness and effectiveness of the alerts. Dreyev aims at replicating the ability of a smart passenger who steps into a car driven by an unknown person to quickly a) assess behavior of the driver, b) evaluate the driver's ability to pay adequate attention to the road, c) predict whether the driver has experience and skills to cope with unexpected events, d) decide whether and how to interact with the driver, depending on severity and urgency. While such copilot role may come naturally to most humans, it is extremely difficult to do for a machine. At dreyev, we believe that augmented intelligence will allow machines to know when we need help, understand what kind of help and level of escalation needed, to interact with us naturally, as our best friends.
Driving is a complex skill and equally complex is the ability to evaluate whether a driver is skilled and attentive enough to manage every day driving situations in the safest possible way. We aim at replicating the important role of a driving coach or a co-pilot, using novel algorithms to integrate machine vision, connected vehicles, smart digital assistants and artificial intelligence. The biggest challenge? Doing it in real time while using an inexpensive computing platform. Highly innovative and challenging are 1. creation of detailed situational contexts 2. models of human-computer interactions 3. evaluation of driving risk and attention level in real-time using multiple visual clues to provide immediate guidance 4. evaluation of the driver’s response to the feedback to adapt communication strategies as needed. Complementary differentiators are the ability to perform driver identity and demographic verification to enable automated selection and tailoring of guidance and monitoring profiles to individual preferences, skills and experience. Other challenges • difficult business model in low motor premium markets (e.g. India, China) • perception of privacy violation* by personal drivers (not the case with drivers on contract, e.g. fleet drivers) • backlash against potential customer tracking by insurance • need for higher-skilled data scientists to analyze and process statistical data about driver behaviors • visual processing of a driver's face and eyes for accurate head pose detection, eye gaze tracking, eyelid closing patterns in highly variable illumination conditions is far from trivial • need for sophisticated Machine Learning applied to a variety of signals to determine safety margins in all possible driving conditions is another significant barrier of entry. • need for specialized technology, patents, and highly specialized experts in areas like Deep Learning applied to Machine Vision, Deep Learning applied to Human-Machine interaction (evaluation of effectiveness of messaging, responsiveness analysis). Regulations will help with the adoption process. Telematics devices have long been in use in commercial vehicles, but the 2018 Electronic Logging Device mandate pushed adoption to a much higher level. Now several US fleet management companies are starting to mandate the use smart video recording devices (VEDR) as safety devices in all their vehicles. Bonus features include the ability to identify risky behaviors of drivers in real time, with immediate feedback to the driver and to the fleet manager. EU is just about to create a mandate for drunken, drowsy and advanced distracted driving warning in new EU cars. The proposal has been worked on over the last 4 years, approved by a vast majority last week, waiting for the final decision by mid March 2019. It will require many new EU cars to implement the new functions by 2022, and all new vehicles by 2024. http://www.europarl.europa.eu/news/en/press-room/20190220IPR27656/safer-roads-more-life-saving-technology-to-be-mandatory-in-vehicles We expect that to become a big market pull for solutions like dreyev. Feasibility Economic Feasibility is determined by the flexible business model, assuming competitive pricing ($300-$350 per device, $30 monthly service fee) in the commercial vehicle market. Usability: Easy to install (box attached to rearview mirror), operates via hands-free, eyes-free communication, effectively timed and personalized. User testing conducted by our subcontractor Westat will help further adjust user experience factors and perfect usability.
Years ago upon landing my first big job at IBM Watson, I faced a 3 hour daily commute. One morning, after a short night sleep, I had almost reached my destination, when just before taking the exit I fell asleep behind the wheel and drove into the ditch. Fortunately the turbulence woke me up and a shot of adrenaline saved me from an even worse crash. Back then I could only dream of a device that would warn me of a danger, or talk to me to keep me alert. Today this has become a reality with dreyev, an in-vehicle intelligent system designed to prevent drowsy and distracted crashes. Health and Well-being Dreyev AI co-driver will have a significant impact on health and well-being in a number of ways, by (1) preventing fatal crashes and saving lives. (2) Reducing severity of accidents, which will lead to shorter recoveries, better outcomes, decreased number of chronic conditions resulting from auto accidents, and therefore fewer cases where chronic pain may lead to addiction to opioid pain relievers, and the overall better quality of life upon recovery from an auto accident. (3) Reducing the stress associated with driving long distances, when fatigue and drowsiness may lead to a crash. (4) Improving quality of live of elderly drivers, who may continue to drive longer and safer thanks to dreyev AI co-pilot that will guide their attention to pivotal moments on the road. (5) Reducing the stress family members experience when their loved ones drive long distances alone or when they have a teen or an elderly driver in the family. Broad vision: Medium-term: reduce the number of auto crashes by 25% by year 2025. The next two decades will see the growth of partially self-driving cars mixed in traffic with traditional vehicles. That will require drivers to pay attention to the road, to be ready to regain control of the car when needed, and will require self-driving cars to understand the attention level of the driver, to decide when they can safely transition control to the driver. The same technology developed for today’s aftermarket scenario will be applicable to self-driving vehicles. Long-term: create AI systems that learn how to effectively communicate with human beings, by interpreting their behavior, communication style, gestures, habits and adapting the interactions based on human responses. Time has come and technology exists to create machines capable of cooperating with humans by using human-style communications, ability to guess intentions, even emotions, as humans do effortlessly.