Road to Autonomy

— 17 May 2017 —

“GOOGLE SELF-DRIVING CAR” © LOKAN SARDARI | FLICKR | CREATIVE COMMONS LICENSE

Introduction & Background

Fully-autonomous vehicles (AV) are poised to become one of the most disruptive technologies of the century — affecting industries and jobs from transportation to healthcare. The implications of AVs for human quality of life are also wide-ranging from the magnitude of our impact on the environment to how much time we spend in traffic. Functions of the automobile have been progressively automated since the 1920s for both convenience and utility, and concerted efforts to develop a self-driving vehicle came to fruition as early as the 1980s [1]. However, only in the last three years has engineering reached a threshold to permit testing of AVs on public roads. In this post, we evaluate the current status of the rapidly-developing AV technology at three relevant scales (global, national, local), using publicly-available data and the geospatial analysis platform CARTO.

The advantages of autonomous transportation and the challenges its integration poses are being undertaken as a worthwhile pursuit for a variety of reasons. Proponents of AVs cite increased traffic safety, decreased traffic congestion, reduced parking space needs, increased carsharing use, and decreased heathlcare, insurance, and law enforcement costs [2] among the chief benefits. However, several obstacles remain in the widespread adaptation of the technology such as software reliability, sensor cost, computer learning sophistication, and public opinion/support… in addition to drawbacks like job loss and cybersecurity-related transportation risks [7].

Nevertheless, recent breakthroughs in geospatial technology, computer processing, and auto manufacturing have culminated to force government regulators to quickly respond to advances. In 2014, the Society of Automotive Engineers (SAE) International created a taxonomy system to define successive stages of automation from “Level 0” wherein automotive systesms and sensors provide feedback to the human driver but have no autonomous control of the vehicle, to “Level 5” where no human intervention is required whatsoever and technically controls like a steering wheel or foot pedals are optional [3]. The United States National Highway Traffic Safety Administration has adopted these definitions as well. Technology has been largely outpacing government regulation and leaving automakers and technology companies searching for the most supportive environments for developing AVs [4]. Currently, however the US maintains its position as the hotspot for AV tech.

Methods & Results

Internationally, only 10 countries currently allow public tesing of AVs in some capacity (ranging from auto manufacturer testing in the United States to a fleet of automated taxis in Singapore) [5]. We visualized these countries allowing AV testing in the Global AV Map using semi-transparent cyan polygon fills. The queries associated with this map are listed at the end of the posting [A]. The 10 countries were China, France, Italy, Japan, The Republic of Korea, Sweden, the United Kingdom, and the Untied States.

GLOBAL AV MAP

In the United States and as of this writing, 15 states have authorized the operation of AVs in some capacity since 2011. However, 33 states in total have introduced legislature in some form related to testing. We visualized these states in the National AV Map using road data to provide an illustration of public areas AV are allowed. The queries associated with this map are listed at the end of the posting [B]. The 15 states were Alabama, Arkansas, Arizona, California, Florida, Louisiana, Massachusetts, Michigan, Nevada, New York, North Dakota, Pennsylvania, Tennessee, Utah, and Virginia — plus the District of Columbia [6].

NATIONAL (UNITED STATES) AV MAP

California (and recently Michigan) is the primary hotspot for AV development in the US, largely due to the concentration of technology companies in Silicon Valley in conjunction with the state’s active and prominent role in regulating auto manufacturers. We wanted to know where the hotspots for testing within the states were and visualized these cites in the Local AV Map using accident data from CA DMV. The number of accidents per city is a proxy for how much testing was performed and should be interpreted loosely (see discussion below). Interestingly but perhaps unsurprising, the San Francisco Bay Area contained the only cities or counties reporting AV-related accidents. The queries associated with this map are listed at the end of the posting [C]. The 5 cities were Los Altos, Mountain View, Palo Alto, San Francisco, and Sunnyvale which encompass 2 counties: San Francisco and Santa Clara [8]. It is important to note most of the accident reports detail “minor” collision incidences.

LOCAL (CALIFORNIA) AV MAP

Discussion & Conclusion

While we used AV accidents per city to quantify the hotspots of testing activity in California because it is some of the only location-based reporting available (i.e., required) in the state or country, this isn’t necessarily very representative of the status of the technology. A more informative metric often discussed is “disengagement reports” required by the state, where AV permitees (i.e., the 21 leading auto or tech companies developing AVs) are required to report the number of times a human driver had to intervene in testing per number of miles driven autonomously [9]. While these are again problematic because the level of detail in companies’ reports vary widely due to competitive pressure limiting information sharing, viewed over time this indicator speaks to the rapid progression of AV technology and particularly which companies are leading the field.

Overall in 2015, AVs drove an average of 183 miles before a disengagement event [9]. In 2016, that number increased to 250 miles without human driver intervention [10]. However, number of miles driven, types of roads driven, and “disengagement success” vary widely within those figures from 0 miles for manufacturers only testing on private facilities, to Waymo (Google) by far leading the field with ca. 1255 miles per disengagement in 2015 and ca. 5125 miles per disengagement in 2016. Recently, it seems AV competition will play out in rather dramatic Silicon Valley fashion with Waymo publicly suing rival Uber for calculated theft of trade secrets. Given that current proof-of-concepts have only attained “Level 3” automation status as of this writing — stakes will remain high in the race to develop and deploy the next global paradigm shift in transportation infrastructure.

Appendicies & References

[A]

SELECT name,ST_Union(the_geom_webmercator) 
AS the_geom_webmercator,
ROW_NUMBER()
OVER (ORDER BY name)
AS cartodb_id
FROM globe
WHERE autoveh = 'testing'
GROUP BY name

[B]

SELECT road.*
FROM road, state
WHERE state.autoveh = 'testing'
AND ST_Contains(state.the_geom_webmercator, road.the_geom_webmercator)

[C]

SELECT cali.*
FROM calicity AS cali, uscounties AS counties
WHERE ST_DWithin(cali.the_geom_webmercator, counties.the_geom_webmercator,50)
ORDER BY crashes

[1]

https://www.cmu.edu/news/stories/archives/2015/july/look-ma-no-hands.html

[2]

http://www.businessinsider.com/morgan-stanley-autonomous-cars-trillion-dollars-2014-9

[3]

http://standards.sae.org/j3016_201401/

[4]

http://docs.house.gov/meetings/IF/IF17/20170214/105548/HHRG-115-IF17-Wstate-KarrbergA-20170214.pdf

[5] https://www.rolandberger.com/publications/publication_pdf/roland_berger_index_autonomous_driving_q3_2016_final_e.pdf

[6]

http://www.ncsl.org/research/transportation/autonomous-vehicles-self-driving-vehicles-enacted-legislation.aspx

[7]

http://www.smh.com.au/business/comment-and-analysis/the-downside-of-driverless-cars-20151105-gkro8y.html

[8]

https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/autonomousveh_ol316+

[9]

https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/testing

[10]

https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/disengagement_report_2015

[11]

https://www.dmv.ca.gov/portal/dmv/detail/vr/autonomous/disengagement_report_2016

— — — — — — — — — — — — — — — — — — — — — — — — — — — — — —

*** THIS POST WAS WRITTEN AS A FINAL PROJECT FOR THE WEB MAPPING COURSE IN THE MASTERS OF ENVIRONMENTAL MANAGEMENT PROGRAM AT THE GEOSPATIAL ANALYSIS LAB, UNIVERSITY OF SAN FRANCISCO. ***

One clap, two clap, three clap, forty?

By clapping more or less, you can signal to us which stories really stand out.