Mike Worry, Nuvation CEO, put together a brief slideshow regarding his view of why it should be illegal for humans to drive and his solution to that problem. Here is a summary of that presentation that I have put into an EDN slideshow.
Worry ticks off some positive reasons to create autonomous vehicles:
- 93% of traffic accidents are due to human error
- 108,000 deaths per month worldwide
- 1.2 million preventable deaths per year
- Vehicles are the leading cause of death for Americans aged four to 34
- 2.3 million US drivers and passengers treated in US emeregency rooms in 2009
- Traffic crashes cost Americans $299.5 billion annually
- 2,996 people died in 9/11 (less than die every day by traffic accidents)
- It is estimated that seat belts have saved 255,000 lives in US since 1975
- The National Highway Traffic Safety Administration (NHTSA) estimates that as of January 1, 2009, more than 28,000 people were alive because of frontal airbags, which became common in the 1990s and have been required in new passenger vehicles since the 1999 model year.
Even more reasons to develop the autonomous vehicle:
- Increase capacity of existing roads
- Even in heavy traffic -- only 8% to 12% of road surface is cars
- Self-driving cars could increase road density by 5x
- Four lane highway costs $8 million to $12 million per mile
- Cars would park themselves -- or simply drive away and return when requested
- In congested urban areas, 40% of gas is expended looking for parking
- Some US cities, more than one third of land area is parking lot
- By that capacity increase, traffic goes away
- Parking problems go away
- Improvement in productivity and lifestyle
- Average American commuter spends 250 hours a year commuting
- Average car sits unused for 22 hours of the day
- Move to models of fractional ownership of cars -- or zipcars by the hour
- Cost of a cab fare with convenience of personal ownership
- Savings in fuel efficiency
- Provide mobility to handicapped
- Provide mobility to an aging population
- Open up new ownership models
- DUIs and drunk driving disappear
The basic design of an autonomous vehicle
History of autonomous vehicles
In 1980, a Mercedes-Benz was driven autonomously using vision in up-to 100 km/h -- without any traffic on that road! Then there was the Prometheus Project where a re-engineered W140 S-Class that technically drove almost entirely by itself over 1678 kilometers (1043 miles) from Munich to Copenhagen back in 1995. This was a VITA vehicle -- VITA was an acronym standing for "Vision Information Technology Application," and was essentially a car with autopilot. There were many more such projects to come.
Then in 2005, DARPA had its Grand challenge for 200 km. Five different vehicles drove autonomously with no traffic.
And in 2007, there was the DARPA Urban Challenge in which nine teams drove stunt driver vehicles on conventional roads.
In 2010, Google got involved (What don't they get involved in nowadays?) and drove a car autonomously in California. VisiLab did one better the same year and saw a car drive autonomously from Italy to China during the Visilab Autonomous Challenge.
In 2011 Audi TT raced up Pike's Peak in 27 minutes! Just 12 minutes slower than a person driving the same route.
In 2012 we saw the first autonomous car license in Nevada. Google in May and then Audi in October -- at a cost of about $1 million you can talk on your phone, text, but no drinking!, says Worry.
How the National Highway Traffic Safety Administration (NHTSA) classifies autonomy
- Level 0: No Automation -- Driver in control, including cars with warning systems, but where there is not automation.
- Level 1: Function-specific automation -- particular sub-systems such as cruise control, lane-keeping, vehicle automates a single aspect, driver still entirely in control
- Level 2: Combined Function Automation -- Two or more inputs are managed together. Combined CC and lane-keeping, traffic-jam assist. Driver takeover with zero warning
- Level 3: Limited Self-Driving Automation -- All functions automated in limited situations. Vehicle monitors and alerts prior to handoff
- Level 4: Full Self-Driving Automation -- Driver completely optional
- Tesla claims it will have full autonomy 90% of the time by 2017
- ABI Research says 2020
- Nissan, Honda, Continental, and Denso Panelists, SAE 2013 World Congress: 2020 to 2025
- Richard Wallace, Centre for Automotive Research @ Stanford: "2025 at the earliest"
- Sven Beiker, Director: Centre for Automotive Research @ Stanford: "20 years from now, we might have completely autonomous vehicles, maybe on limited roads"
- Jürgen Leohold, Executive Director VW Research Group: 50 years
- Steven Shladover, Researcher: California Partners for Advanced Transportation Technology: "I don't expect to see the fully-automated, autonomous vehicle out on the road in the lifetime of anyone in this room"
New Technologies to make it happen in 2013 through 2014
- Sensors and computation are getting smaller, more hidden
- Most car companies have demonstration vehicles working on known routes in perfect conditions
- Expectations are being tempered, gradual rollout, expensive at firstOEM prototypes
Mercedes technology 2013 through 2014
- Mercedes demonstrated vision centric autonomous driving
- Close to production sensors only
- At least four cameras + IR, stereo vision for depth information, radar, sonar
- Visual localization robust to lighting changes
- Matching of a feature map created in advance of drive
- Visual signage, lane detection
- Known locations, only confirming
- Specific route predefined
- Only local online planning to avoid obstacles, lane level route predefined
Problems still loom
- Hardest problem for computers is perception of the external environment
- A human can use inferential thinking from 16 years of living, combined stored memories and sensory input to interpret events and anticipate likely scenarios.
- E.g., A soccer ball rolling into the road may suggest a kid is about to run into the street
- E.g., Distinguishing mass and density of obstacles. A steel block must be avoided. A flapping shopping bag can be driven over.
Next page: Mercedes technology 2013/14
Autonomous vehicle focus/green power and transportation
- Substantial electronics growth opportunity in autonomous vehicle space
- Nuvation has a four-year research agreement with the University of Waterloo to research new product opportunities for autonomous vehicles
- Identification of hardware and algorithmic limitations/solutions
- Coordinated Platooning
- Coordinated Emergency Braking
- Vehicle Motion Estimation
- Static and Dynamic Object Detection/Tracking
- Reference designs for system components
- Development of a convenient testbed
- Integration of Communication and Sensor Data
Autonomous vehicle transportation by Nuvation-Discofish
See my EDN article Can engineering be fun?: Nuvation and 'Burning Man.'
Worry and his team designed and built Discofish in the after-hours of their business day of creating innovative customer designs.
- DiscoFish: Nuvation's testbed to build domain knowledge of autonomous vehicle driving systems
- Features drive-by-wire capabilities, Garmin 17HVS GPS, and SICK LMS111 LIDAR
- Partnership with Waterloo Autonomous Vehicle Laboratory to spearhead research in perception, planning, and coordination of autonomous vehicles