Jeffrey Funk

Retired from

National University of Singapore,

Hitotsubashi University, Kobe University, Penn

State, Carnegie Mellon, University of Michigan

For information on other technologies, see http://www.slideshare.net/Funk98/presentations

The First Cars were Implemented in a Constrained Environment

Paved roads were created for autos

Highways were created for fast moving autos

Special entry points

Horses, bicycles, and old vehicles aren’t allowed

Fences prevent entry by animals and children at other points

These paved roads and high-ways reduce complexity of driving and thus increase safety

Other Technologies also Implemented in Constrained Environment

Planes use airports and special flight corridors

Ships uses ports and special corridors within ports

IT uses standards to simplify design

Interface standards exist for most products

Compatibility may emerge later (e.g., Wintel and Apple computers)

Shouldn’t We “Constrain” the Environment for Driverless Vehicles?

Won’t allowing them on all roads and all parking lots be dangerous?

Without constraints, AVs must handle many contingencies

Children run onto road

Cars run out of gas or break down

Street or traffic lights stop working

Chaos of parking lots

Bad or Unusual Weather Provides Other Reasons for Constraints

Difficult situations Dark, Raining

Snowing

Foggy, Windy

It will take many years for driverless vehicles to handle all situations

Would you drive next to driverless truck on snowy day?

Without Constraints, the Benefits from AVs are very Small

Drivers can do something else while AV is self-driving Read, watch videos

Is this a large benefit?

Governments may allow driver to be eliminated Reduces cost of taxis

Increases capacity of taxis

Is this a large benefit and when might governments allow these changes?

Shouldn’t we be looking for bigger benefits?

Shouldn’t we be Looking for Larger Benefits

Can we move these vehicles at 60 MPH? Reducing travel time is

potentially big benefit

When roads are completely filled with driverless vehicles Inter-vehicle distances can be

reduced

Traffic signals can be eliminated

Both enable higher capacity roads, perhaps enabling roads to be used for something else

25% of space in Los Angeles is for roads and parking lots

City Percentage Devoted to Streets

Street Area (square feet) Per Capita

New York 30% 345

Newark 16% 257

San Francisco 26% 441

Chicago 24% 424

Philadelphia 19% 365

St. Louis 25% 609

Pittsburgh 18% 455

Cleveland 17% 416

Miami 24% 778

Milwaukee 20% 724

Cincinnati 13% 573

Los Angeles 14% 741

Atlanta 15% 1,120

Houston 13% 1.585

Dallas 13% 1,575

Portion of Land Devoted to StreetsSource: John R. Meyer and Jose A. Gomez-Ibanez, Autos, Transit, and Cities, Twentieth

Century Fund Report (Cambridge: Harvard University Press, 1981).

Rank City Parking Area* Divided by Land Area

1 Los Angeles 81%

2 Melbourne 76%3 Adelaide 73%

4 Houston 57%

5 Detroit 56%

6 Washington, D.C. 54%

7 Brisbane 52%

8 Calgary 47%

9 Portland 46%

10 Brussels 45%

Land for Parking in Urban AreasSource: Michael Manville and Donald Shoup, “People, Parking, and Cities,” Journal of Urban Planning and Development, Vol.

131, No. 4, December 2005, pp. 233-245

* Includes all levels of all parking garages

The Bottom Line

Safety problems are large as long as both AVs and conventional vehicles are interacting on roads and in parking lots

Elimination of driver and driver’s seat is small benefit

The benefits from driverless vehicles don’t become large until all vehicles on a road (or lane of road) are driverless vehicles This should be the goal of driverless vehicles

Cities can charge users for access to roads (or lanes) dedicated to AVs

New revenue source for cities, which can be used for many things

Constraining the environment can increase safety and reduce the cost of the vehicles

What Might These “Autonomous Roads” (or Lanes in Roads) be Like?

Vehicles are Controlled by Wireless Communication Technologies on Dedicated Roads

Cars are checked for autonomous capability when they enter a dedicated road

Route plans are checked and integrated with other route plans

Improvements in computer processing power facilitate checking and integrating

Much of these calculations would be done in secure private cloud

Other Simple Solutions that Provide Additional Safety

Magnets and RFID tags can be embedded in highways to help control vehicles

They create an invisible railway

Estimated cost in Singapore <200M SGD for magnets <110M SGD for RFID Very cheap, less than 2SGD

per vehicle

Dedicated Roads Lead to Higher Capacity Roads

Dedicated Roads Lead to Fewer Delays at Traffic Signals

Roads dedicated to AVs can have higher speeds and

thus higher Fuel Efficiencies (lower carbon emissions)

Can we move these

cars at 30MPH or faster?

Latency is Key Issue but it is Still Falling

Expected to fall below 0.1 milliseconds with wireless 5G services that will be implemented by early 2020s Jones R 2015. Telecom’s Next Goal: Defining 5G, Wall Street Journal, March 9.

http://www.wsj.com/articles/telecom-industry-bets-on-5g-1425895320

Could AVs become the main market for cellular 5G services?

Processing is done in cloud and the cost of these cloud services continues to fall

Falling latency requires better IT, but this keeps occurring through Moore’s Law

Improvements in Latency (delay times in

milliseconds) Enable Centralized Control of Vehicles

High Processing Capability is Needed to Control Vehicles

Improvements in Integrated Circuits and Computers Enable this Processing Power

Processing power for 100 km road by vehicle inflow and reaction times

(Several thousands PCs)

Many of the Computer Calculations (price per car)

Would be Done in the Cloud

Moore’s Law Drives Reductions in Cloud

Computing Services (price per car)

Let’s Design “Autonomous Roads” for AVs

Dedicate roads or lanes in roads to AVs

Over time increase number of roads (or lanes) that are dedicated to AVs

This would

Increase safety of AVs, while increasing benefits from AVs

And reducing cost of AVs

Cost of AVs is already falling rapidly (see subsequent slides)

Emphasizing wireless control will reduce necessary on-car capabilities and thus cost of AVs

<$5,000 per car is possible

Capabilities can be embedded in module that can be added to existing vehicles

Begin with Highways

Benefit from higher density of cars per area, all fast moving

Eliminate some highways (or lanes) since autonomous highways have more capacity

Then Transform Surface Streets Higher capacity of

autonomous roads enables some roads to be used for other purposes

Autonomous roads can be surrounded by fences and perhaps roofs, thus enabling parks or other facilities to be constructed on top of them

Cost of Autonomous Vehicles (Google Car) Falls as Improvements

in Lasers and Other “Components” Occur

Source: Wired Magazine, http://www.wired.com/magazine/2012/01/ff_autonomouscars/3/

Better Lasers, Camera chips, MEMS, ICs, GPS Are Making these

Vehicles Economically Feasible1 Radar: triggers alert when something

is in blind spot

2 Lane-keeping: Cameras recognize lane

markings by spotting contrast between road

surface and boundary lines

3 LIDAR: Light Detection and Ranging

system depends on 64 lasers, spinning at

upwards of 900 rpm, to generate a 360-

degree view

4 Infrared Camera: camera detects

objects

5 Stereo Vision: two cameras build a

real-time 3-D image of the road ahead

6 GPS/Inertial Measurement: tells us

location on map

7 Wheel Encoder: wheel-mounted

sensors measure wheel velocity

ICs interpret and act on this data

Falling Cost of Autonomous Vehicles

Cost of “Google Car” was $150,000 in 2012

mostly for electronic components

about $70,000 for LIDAR from Velodyne

Current rates of improvement are 30%-40%

If costs drop 25% a year, cost of electronics will drop by 90% in ten years

May be evolutionary move towards AVs as Sensors are incorporated into existing vehicles http://www.ti.com/ww/en/analog/car-of-

the-future/?DCMP=gma-tra-carofthefuture-en&HQS=carofthefuture-bs-en

But many of these costs have dropped faster than this calculation

Velodyne offers low-cost LIDAR for $8,000

http://www.theguardian.com/technology/2013/jun/02/autonomous-cars-expensive-google-

http://www.wsj.com/articles/continental-buys-sensor-technology-for-self-driving-cars-1457042039

Cost of Self-Driving Car Feature Self-Driving Car Volume Forecast

Other Cost (and Volume) Estimates for AVs

• Cost is key hurdle of Google’s self driving car

• Cost ~ $200,000 to build in 2014

• By 2015, cost reduced to $50,000

• Further reduction as technology matures and volume increase

• Look out for cost to reach $7000. Will lead to rapid adoption

Wireless Control Enables Much Cheaper AVs

Inexpensive modules (<$5,000) can be produced using wireless and other integrated circuits

In addition to new vehicles, existing vehicles can be retrofitted with these modules

No need for LIDAR because of constrained environment

Lower costs enable faster diffusion

Faster diffusion enables faster implementation of roads dedicated to AVs

Multiple Scenarios Can be Pursued Simultaneously

Scenario emphasized in these slides is design autonomous roads for AVs

This can be pursued even as mixed road scenario is pursued

High-end AVs are sold and they are used on roads with manually driven cars

These AVs will likely require divers for many years

But if they are successful, the drivers and the driving wheel may be eliminated, thus promoting the diffusion of these high-end AVs

Once these AVs have diffused, cities might pursue fully autonomous roads

Many Challenges for Autonomous Roads

Need a good architecture and conceptual design for both system and vehicle modules

Need cellular infrastructure suppliers to work with automobile companies, component suppliers, and cities to design and test systems

Tests would be required under many types of weather situations

The goal should be operational systems by 2025, just as 5G has begun to diffuse

Many Challenges (2)

Changeover from existing to autonomous roads will be difficult

Will enough people be willing to purchase modules to justify fast changeover?

Or will autonomous roads be under utilized for many years, thus wasting scarce resource of land?

What about people who don’t buy modules?

If they can’t use specific highway, what can they do?

They must be given viable alternatives

Can we offer them public transport or inexpensive multiple passenger ride sharing services?

Will they accept change or fight it?

Many Challenges (3)

Alternatively, can we begin with lanes in roads, rather than entire roads?

Dedicate one lane to AVs

This would allow gradual switch from fully manual to fully autonomous road

One problem:

when highways are crowded, only the AV lane will be moving

How would an AV exit in this situation?

Would all the AVs have to stop for an AV to exit?

Summary

AVs are quickly becoming cheaper

But their costs will remain high and their benefits low until we have fully autonomous roads

Developing these roads should be the goal of AVs

For naysayers, technologies have always been initially implemented in constrained environments

AVs should also be implemented in this way in order

increase safety

reduce costs of implementation

increase benefits from implementation