Seeing Around Corners: Alibaba Rethinks Automated Driving

How Alibaba is making AI-navigation for driverless cars a part of the road itself

Image for post
Image for post

In the search for truly dependable driverless vehicles, a great deal of effort has gone into developing onboard AI systems that can ensure that cars recognize and react to their surroundings without fail. More recently, though, the pursuit of safety has turned outward, as developers increasingly ask how roads and surveillance infrastructures can be harnessed to better guide automated travel.

Now, with a cloud-based solution that puts cars and smart surveillance systems in contact, the Alibaba Group is advancing an approach that gives vehicles a bird’s eye view of potential dangers before they come into range. Built around a series of external monitoring stations that communicate with onboard systems, and helped by a simulation module that gives computers tens of thousands of miles of driving experience, it is already showing how reaction-based driving could soon be a thing of the past.

In his speech at the recent 2018 Yunqi Computing Conference, Alibaba’s head of automatic driving Wang Gang presented his team’s achievement while tracing the historical developments that brought it into focus. Inside, we look at key points from Wang’s presentation and explore how AI that sees around corners is paving the way for a future without blind spots.

Image for post
Image for post
Alibaba’s Wang Gang delivers his address “Automated Driving: Evolving from Single-seat AI Navigation to Collaborative Intelligence” at the 2018 Yunqi Computing Conference

Starting small: From prototypes to single-seaters

Contrary to popular belief that driverless vehicles reflect a recent area of research, the road to today’s frontier stretches back well into the 20th century, marked by considerable distances between developments. Even starting from the first successful prototype (developed at Carnegie Mellon University in 1984), it took another two decades for a field of contestants to answer the first DARPA Grand Challenge in 2004, none of whose models reached the end of the competition’s desert race course before breaking down.

In terms of tangible progress, much of Alibaba’s early work on autonomous navigation began with self-driving single-seat cars. Believing that navigation systems needed to be a part of cars themselves, developers were able to make all-weather positioning systems accurate to within a centimeter for these vehicles. Following this achievement, robust efforts yielded an industry-leading 3D object recognition algorithm, starting a period of two years in which the group’s scientists published 12 thought-leading papers to the CVPR conference. Together these efforts established Alibaba’s competitiveness in driverless innovation and yielded a complete set of automated driving systems to refine through testing and verification.

In testing, the group’s work benefited from the hectic and even chaotic traffic scenarios China’s roadways present, as the problems in any given model quickly become apparent under such conditions. Ultimately, the difficulties of navigating such an environment revealed the limits of a vehicle-centric approach altogether — namely blind spots, depth perception issues, delays in high-precision map updates, and decision-making plans that fail to report information completely. Seen another way, these problems all become issues of cost, as the tools providing even such limited capabilities cost as much as hundreds of thousands of dollars per vehicle.

Without a cost-effective way to eliminate many of the same vulnerabilities inherent in human driving, Alibaba is, at this point, committed to looking for AI solutions beyond individual car systems while continuing to seek an application for the capabilities it had already developed.

Shifting to collaborative intelligence

Rather than an alternative to its onboard AI model, Alibaba’s next wave of efforts sought a complementary technology to assist driverless vehicles while allowing them to remain partially autonomous. Known as collaborative intelligence, the resulting approach greatly reduces the navigation burden on cars by allowing them to take signals from external AI tools built into the physical infrastructure of roadways. Simultaneously, its continued use of onboard navigation systems in the basic challenges of driving enables infrastructure-based tools to focus on large-scale monitoring and coordination activities, making the system as a whole truly collaborative.

To put this concept in action, developers created a network of IntelliSense base stations each able to monitor traffic conditions within a 400-meter radius, placing the stations in strategic positions such as above the corners of intersections. Adding to the benefits of this vantage, each station’s sensors operate from a fixed point in space, eliminating many of the difficulties present in calculating depth and velocity from onboard a moving vehicle. More importantly, the IntelliSense stations were designed with cloud connectivity that enables them to communicate with each other at all times, such that each acts as one eye of a near-omniscient awareness of citywide traffic conditions.

Image for post
Image for post
Console view of operations on an IntelliSense base station

In a final security measure drawing on the Alibaba Group’s extensive work with machine learning, developers trained the navigation systems with simulation footage amounting to tens of thousands of miles of rote driving experience.

In trials, the collaborative intelligence model proved able to prevent accidents that would almost surely occur at the hands of human drivers. Specifically, Alibaba’s team tested vehicles connected to IntelliSense in a 30 kilometer per hour trial of their ability to react to the sudden appearance of a pedestrian at a distance that results in injury in roughly 100 percent of all cases. The driverless vehicle completely inverted its human counterpart’s performance during the trial, stopping in time to prevent injury in every single iteration of the scenario.

Toward a safer, cheaper, driverless future

As well as demonstrating a novel, safety-enhanced approach to AI navigation, collaborative intelligence offers an economical means to a future without the need of drivers. With over 300 million vehicles concentrated in relatively dense clusters of urban roadways, China’s transportation system overwhelmingly favors a conversion to smart roads over a conversion to smart vehicles. As a compromise between the two, collaborative intelligence enables a smarter car to interact with a smarter infrastructure, letting the two do together what neither can yet do alone.

Looking ahead, Alibaba has now partnered with the Highway Department of China’s Ministry of Transportation to launch a collaborative laboratory for enhancing infrastructure through technology, which it hopes will accelerate the adoption of such advances in China’s increasingly connected cities. As Wang Gang concluded in his presentation to the Yunqi Computing Conference, the mandate for such technologies has never been clearer as navigation systems can now ensure a whole new level of vehicular safety for passengers and pedestrians alike.

Alibaba Tech

First hand and in-depth information about Alibaba’s latest technology → Facebook: “Alibaba Tech”. Twitter: “AlibabaTech”.

Written by

First-hand & in-depth information about Alibaba's tech innovation in Artificial Intelligence, Big Data & Computer Engineering. Follow us on Facebook!

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store