Driverless car

A driverless car is a vehicle equipped with an autopilot system, and capable of driving from one point to another without aid from an operator. Driverless passenger car programs include the 800 million EC EUREKA Prometheus Project on autonomous vehicles, the 2getthere passenger vehicles from the Netherlands, the ARGO research project from Italy, the DARPA Grand Challenge from the USA, and Google driverless car.

History

An early representation of the driverless car was Norman Bel Geddes's Futurama exhibit sponsored by General Motors at the 1933 World's Fair, which depicted electric cars powered by circuits embedded in the roadway and controlled by radio.

The history of autonomous vehicles starts in 1977 with the Tsukuba Mechanical Engineering Lab in Japan. On a dedicated, clearly marked course it achieved speeds of up to 30 km/h (20 miles per hour), by tracking white street markers (special hardware was necessary, since commercial computers were much slower than they are today).

In the 1980s a vision-guided Mercedes-Benz robot van, designed by Ernst Dickmanns and his team at the Bundeswehr University of Munich in Munich, Germany, achieved 100 km/h on streets without traffic. Subsequently, the European Commission began funding the 800 million Euro EUREKA Prometheus Project on autonomous vehicles (1987–1995).

Also in the 1980s the DARPA-funded Autonomous Land Vehicle (ALV) in the United States achieved the first road-following demonstration that used laser radar (Environmental Research Institute of Michigan), computer vision (Carnegie Mellon University and SRI), and autonomous robotic control (Carnegie Mellon and Martin Marietta) to control a driverless vehicle up to 30 km/h. In 1987, HRL Laboratories (formerly Hughes Research Labs) demonstrated the first off-road map and sensor-based autonomous navigation on the ALV. The vehicle travelled over 600m at 3 km/h on complex terrain with steep slopes, ravines, large rocks, and vegetation.

In 1994, the twin robot vehicles VaMP and Vita-2 of Daimler-Benz and Ernst Dickmanns of UniBwM drove more than one thousand kilometers on a Paris three-lane highway in standard heavy traffic at speeds up to 130 km/h, albeit semi-autonomously with human interventions. They demonstrated autonomous driving in free lanes, convoy driving, and lane changes left and right with autonomous passing of other cars.

In 1995, Dickmanns´ re-engineered autonomous S-Class Mercedes-Benz took a 1600 km trip from Munich in Bavaria to Copenhagen in Denmark and back, using saccadic computer vision and transputers to react in real time. The robot achieved speeds exceeding 175 km/h on the German Autobahn, with a mean time between human interventions of 9 km, or 95% autonomous driving. Again it drove in traffic, executing manoeuvres to pass other cars. Despite being a research system without emphasis on long distance reliability, it drove up to 158 km without human intervention.

In 1995, the Carnegie Mellon University Navlab project achieved 98.2% autonomous driving on a 5000 km (3000-mile) "No hands across America" trip. This car, however, was semi-autonomous by nature: it used neural networks to control the steering wheel, but throttle and brakes were human-controlled.

From 1996–2001, Alberto Broggi of the University of Parma launched the ARGO Project, which worked on enabling a modified Lancia Thema to follow the normal (painted) lane marks in an unmodified highway. The culmination of the project was a journey of 2,000 km over six days on the motorways of northern Italy dubbed MilleMiglia in Automatico, with an average speed of 90 km/h. 94% of the time the car was in fully automatic mode, with the longest automatic stretch being 54 km. The vehicle had only two black-and-white low-cost video cameras on board, and used stereoscopic vision algorithms to understand its environment, as opposed to the "laser, radar - whatever you need" approach taken by other efforts in the field.

Three US Government funded military efforts known as Demo I (US Army), Demo II (DARPA), and Demo III (US Army), are currently underway. Demo III (2001)demonstrated the ability of unmanned ground vehicles to navigate miles of difficult off-road terrain, avoiding obstacles such as rocks and trees. James Albus at NIST provided the Real-Time Control System which is a hierarchical control system. Not only were individual vehicles controlled (e.g. throttle, steering, and brake), but groups of vehicles had their movements automatically coordinated in response to high level goals.

In 2002, the DARPA Grand Challenge competitions were announced. The 2004 and 2005 DARPA competitions allowed international teams to compete in fully autonomous vehicle races over rough unpaved terrain and in a non-populated suburban setting. The 2007 DARPA challenge, the DARPA urban challenge, involved autonomous cars driving in an urban setting.

In 2008, General Motors stated that they will begin testing driverless cars by 2015, and they could be on the road by 2018 .

In 2010 VisLab ran VIAC, the VisLab Intercontinental Autonomous Challenge, a 13,000 km test run of autonomous vehicles. The four driverless electric vans successfully ended the drive from Italy to China via the arriving at the Shanghai Expo on 28 October.

Recent projects

The work done so far varies significantly in its ambition and its demands in terms of modification of the infrastructure. Broadly, there are three approaches:

Fully autonomous vehicles

Various enhancements to the infrastructure (either an entire area, or specific lanes) to create a self-driving closed system.

"assistance" systems that incrementally remove requirements from the human driver (e.g. improvements to cruise control)

An important concept that cuts across several of the efforts is vehicle platoons. In order to better utilize road-space, vehicles are assembled into ad-hoc train-like "platoons", where the driver (either human or automatic) of the first vehicle makes all decisions for the entire platoon. All other vehicles simply follow the lead of the first vehicle.

]Fully autonomous

Fully autonomous driving requires a car to drive itself to a pre-set target using unmodified infrastructure. The final goal of safe door-to-door transportation in arbitrary environments is not yet reached though.

Vehicles for paved roads

Google driverless car, with a test fleet of autonomous vehicles that by October 2010 have driven 140,000 miles (230,000 km) without any incidents.

The 800 million Euro EUREKA Prometheus Project on autonomous vehicles (1987–1995). Among its culmination points were the twin robot vehicles VITA-2 and VaMP of Daimler-Benz and Ernst Dickmanns, driving long distances in heavy traffic (see #History above).

The VIAC Challenge, in which 4 vehicles drove from Italy to China on a 13,000 kilometres (8,100 mi) trip with only limited occasions intervene by human, such as in the Moscow traffic jams and when passing toll stations. This is the longest-ever trip by an unmanned vehicle.

The third competition of the DARPA Grand Challenge held in November 2007. 53 teams qualified initially, but after a series of qualifying rounds, only eleven teams entered the final race. Of these, six teams completed navigating through the non-populated urban environment, and the Carnegie Mellon University team won the $2 million prize.

The ARGO vehicle (see #History above) is the predecessor of the BRAiVE vehicle, both from the University of Parma's VisLab. Argo was developed in 1996 and demonstrated to the world in 1998; BRAiVE was developed in 2008 and firstly demonstrated in 2009 at the IEEE IV conference in Xi'an, China.

Stanford Racing Team's Junior car is an autonomous driverless car for paved roads. It is intended for civilian use.

The Volkswagen Golf GTI 53+1 is a modified Volkswagen Golf GTI capable of autonomous driving. The Golf GTI 53+1 features a implemented system that can be integrated into any car. This system is based around the MicroAutoBox from dSpace. This, as it was intended to test VW hardware without a human driver (for consistent test results).

The Audi TTS Pikes Peak is a modified Audi TTS, working entirely on GPS, and thus without additional sensors. The car was designed by Burkhard Huhnke of Volkswagen Research.

Stadtpilot, Technical University Braunschweig

AutoNOMOS - part of the Artificial Intelligence Group of the Freie Universität Berlin

Free-ranging vehicles

There are three clusters of activity relating to free-ranging off-road cars. Some of these projects are military-oriented.

US military DARPA Grand Challenge

Main article: DARPA Grand Challenge

The US Department of Defense announced on the July 30, 2002 a "Grand Challenge", for US-based teams to produce a vehicle that could autonomously navigate and reach a target in the desert of the south western USA.

In March 2004, the first competition was held, for a prize-money of $1 million. Not one of the 25 entrants completed the course. However, in the second competition held in October 2005 five different teams completed the 135-mile (217 km) course, and the Stanford University team won the $2 million prize.

November 3rd, 2007, the third competition was held and $3.5 million dollar in cash prizes, trophies and medals were awarded. Six driverless vehicles were able to complete the 55 miles (89 km) of urban traffic in the 2007 DARPA Urban Challenge rally style race. 1st Place - Tartan Racing, Pittsburgh, PA; 2nd Place - Stanford Racing Team, Stanford, CA; 3rd Place - Victor Tango, Blacksburg, VA.

European Land-Robot Trial (ELROB)

The German Department of Defense held an exhibition trade show (ELROB) for demonstrating automated vehicles in May 2006. The event included various military automated and remotely-operated robots, for various military uses. Some of the systems on display could be ordered and implemented immediately. In August 2007 a civilian version of the event was held in Switzerland.

The Smart team from Switzerland presented "a Vehicle for Autonomous Navigation and Mapping in Outdoor Environments". For pictures of their ELROB demo, see this.

The Israeli Military-Industrial Complex

As a followup from its success with Unmanned Combat Air Vehicles, and following the construction of the Israeli West Bank barrier there has been significant interest in developing a fully automated border-patrol vehicle. Two projects, by Elbit Systems and Israel Aircraft Industries are both based on the locally-produced Armored "Tomcar" and have the specific purpose of patrolling barrier fences against intrusions.

The "SciAutonics II" team in the 2004 DARPA Challenge used Elbit's version of the Tomcar.

Pre-built infrastructure

The following projects were conceived as practical attempts to use available technology in an incremental manner to solve specific problems, like transport within a defined campus area, or driving along a stretch of motorway. The technologies are proven, and the main barrier to widespread implementation is the cost of deploying the infrastructure. Such systems already function in many airports, on railroads, and in some European towns.

Dual mode transit - monorail

There is a family of projects, all currently still at the experimental stage, that would combine the flexibility of a private automobile with the benefits of a monorail system. The idea is that privately-owned cars would be built with the ability to dock themselves onto a public monorail system, where they become part of a centrally managed, fully computerized transport system—more akin to a driverless train system (as already found in airports) than to a driverless car. This idea is also known as Dual mode transit. (See also Personal rapid transit for another concept along those lines, for purely public transport.)

Groups working on this concept are:

RUF (Denmark)

BiWay (UK)

ATN (New Zealand)

TriTrack (Texas, United States)

Automated highway systems

Automated highway systems (AHS) are an effort to construct special lanes on existing highways that would be equipped with magnets or other infrastructure to allow vehicles to stay in the center of the lane, while communicating with other vehicles (and with a central system) to avoid collision and manage traffic. Like the dual-mode monorail, the idea is that cars remain private and independent, and just use the AHS system as a quick way to move along designated routes. AHS allows specially equipped cars to join the system using special 'acceleration lanes' and to leave through 'deceleration lanes'. When leaving the system each car verifies that its driver is ready to take control of the vehicle, and if that is not the case, the system parks the car safely in a predesignated area.

Some implementations use radar to avoid collisions and coordinate speed.

One example that uses this implementation is the AHS demo of 1997 near San Diego, sponsored by the US government, in coordination with the State of California and Carnegie Mellon University. The test site is a 12-kilometer, high-occupancy-vehicle (HOV) segment of Interstate 15, 16 kilometers north of downtown San Diego. The event generated much press coverage.

This concerted effort by the US government seems to have been pretty much abandoned because of social and political forces, above all else the desire to create a less futuristic and more marketable solution.

As of 2007, a three-year project is underway to allow robot controlled vehicles, including buses and trucks, to use a special lane along 20 Interstate 805. The intention is to allow the vehicles to travel at shorter following distances and thereby allow more vehicles to use the lanes. The vehicles will still have drivers since they need to enter and exit the special lanes. The system is being designed by Swoop Technology, based in San Diego county.

Free-ranging on grid

Frog Navigation Systems (the Netherlands) applies the FROG (free-ranging on grid) technology. The technology consists of a combination of autonomous vehicles and a supervisory central system. The company's purpose-built electric vehicles locate themselves using odometry readings, recalibrating themselves occasionally using a "maze" of magnets embedded in the environment, and GPS. The cars avoid collisions with obstacles located in the environment using laser (long range) and ultra-sonic (short-range) sensors.

The vehicles are completely autonomous and plan their own routes from A to B. The supervisory system merely administers the operations and directs traffic where required. The system has been applied both indoors and outdoors, and in environments where 100+ automated vehicles are operational (container port). At this time the system is not suited yet for running the sheer number of vehicles encountered in urban settings. The company also has no intention of developing such technology at this time.

The FROG system is deployed for industrial purposes in factory sites, and is marketed as a pilot public transport system in the city of Capelle aan den IJssel by its subsidiary 2getthere. This system experienced an accident that proved to be caused by a Human error.

Frog Navigation Systems is one of few fully commercial companies in this field.

Driver-assistance

Though these products and projects do not aim explicitly to create a fully autonomous car, they are seen as incremental stepping-stones in that direction. Many of the technologies detailed below will probably serve as components of any future driverless car — meanwhile they are being marketed as gadgets that assist human drivers in one way or another. This approach is slowly trickling into standard cars (e.g. improvements to cruise control).

Driver-assistance mechanisms are of several distinct types, sensorial-informative, actuation-corrective, and systemic.

Sensorial-informative

These systems warn or inform the driver about events that may have passed unnoticed, such as

Lane Departure Warning System (LDWS), for example from Iteris or Mobileye N.V.

Rear-view alarm, to detect obstacles behind.

Visibility aids for the driver, to cover blind spots and enhanced vision systems such as radar, wireless vehicle safety communications and night vision.

Infrastructure-based, driver warning/information-giving systems, such as those developed by the Japanese government

Actuation-corrective

These systems modify the driver's instructions so as to execute them in a more effective way, for example the most widely deployed system of this type is ABS; conversely power steering is not a control mechanism, but just a convenience - it is not involved in decision making.

Anti-lock braking system (ABS) (also Emergency Braking Assistance (EBA), often coupled with Electronic brake force distribution (EBD), which prevents the brakes from locking and losing traction while braking. This shortens stopping distances in most cases and, more importantly, allows the driver to steer the vehicle while braking.

Traction control system (TCS) actuates brakes or reduces throttle to restore traction if driven wheels begin to spin.

Four wheel drive (AWD) with a centre differential. Distributing power to all four wheels lessens the chances of wheel spin. It also suffers less from oversteer and understeer.

Electronic Stability Control (ESC) (also known for Mercedes-Benz proprietary Electronic Stability Program (ESP), Acceleration Slip Regulation (ASR) and Electronic differential lock (EDL)). Uses various sensors to intervene when the car senses a possible loss of control. The car's control unit can reduce power from the engine and even apply the brakes on individual wheels to prevent the car from understeering or oversteering.

Dynamic steering response (DSR) corrects the rate of power steering system to adapt it to vehicle's speed and road conditions.

A review of the overall "feel" to actuation-correction in a Jaguar XK convertible.

Driver-assistance preview from Popular Science (dated 2004).

Note: The electronic differential lock (EDL) employed by Volkswagen is not - as the name suggests - a differential lock at all. Sensors monitor wheel speeds, and if one is rotating substantially faster than the other (i.e. slipping) the EDL system momentarily brakes it. This effectively transfers all the power to the other wheel.

Systemic

Automatic parking: e.g. technology from Ford or Toyota selling for $700, with a 70% take-up rate. The Lexus LS can park itself (parallel/reverse) via the 'Advanced Parking Guidance System' – though only controlling the steering.

Follow another car on a motorway ("Enhanced" or "adaptive" cruise control), like The Ford or Vauxhall(GM).

Nissan's "Distance Control assist"

Dead Man's Switch; there is a move to introduce deadman's braking into automotive application, primarily heavy vehicles, and there may also be a need to add penalty switches to cruise controls.

See also Safety Features.

Existing and missing technologies

In order to drive a car, a system would need to:

Understand its immediate environment (Sensors)

Know where it is and where it wants to go (Navigation)

Find its way in the traffic (Motion planning)

Operate the mechanics of the vehicle (Actuation)

Arguably, 2½ of these problems are already solved: Navigation and Actuation completely, and Sensors partially, but improving fast. The main unsolved part is the motion planning.

Sensors

Sensors employed in driverless cars vary from the minimalist ARGO project's monochrome stereoscopy to Mobileye's inter-modal (video, infra-red, laser, radar) approach. The minimalist approach imitates the human situation most closely, while the multi-modal approach is "greedy" in the sense that it seeks to obtain as much information as is possible by current technology, even at the occasional cost of one car's detection system interfering with another's.

Mobileye N.V. is a technology company that focuses on the development of vision-based Advanced Driver Assistance Systems (ADAS) providing warnings for collision prevention and mitigation. Mobileye offers a wide range of driver safety solutions combining artificial vision image processing, multiple technological applications and information technology. Mobileye's vehicle detection systems, are currently only used for driver assistance, but are eminently suitable for a full-fledged driverless car. This video demonstrates the capabilities of the system: all pedestrians, cars, motorbikes etc. are clearly displayed in video, with a frame around them and the distance between "our" car and the object observed. The system also detects the objects' motion (direction and speed) and can so calculate relative speeds, and predict collisions.

Japanese infra-red article

some things from the DARPA challenge....

Road-sign recognition.

Navigation

The ability to plot a route from where the vehicle is to where the user wants to be has been available for several years. These systems, based on the US military's Global Positioning System are now available as standard car fittings, and use satellite transmissions to ascertain the current location, and an on-board street database to derive a route to the target. The more sophisticated systems also receive radio updates on road blockages, and adapt accordingly. There are also sensors that greatly affect the whole nature of it.

See the main article on Automotive navigation systems.

Motion planning

This is current research problem. See the main article on the subject Motion planning.

Control of vehicle

As automotive technology matures, more and more functions of the underlying engine, gearbox etc. are no longer directly controlled by the driver by mechanical means, but rather via a computer, which receives instructions from the driver as inputs and delivers the desired effect by means of electronic throttle control, and other drive-by-wire elements. Therefore, the technology for a computer to control all aspects of a vehicle is well understood.

Work done in simulation

While developing control systems for real cars is very costly in terms of both time and money, much work can be done in simulations of various complexity. Systems developed using simpler simulators can gradually be transferred to more complex simulators, and in the end to real vehicles. Some approaches that rely on learning requires starting in a simulation to be viable at all, for example evolutionary robotics approaches - see this example.

Social impact

Driverless cars may yield advantages of increasing roadway capacity by reducing the distances between cars, reduce congestion by efficiently controlling the flow of traffic, and increase safety by eliminating driver error.

According to urban designer and futurist Michael E. Arth, driverless electric vehicles—in conjunction with the increased use of virtual reality for work, travel, and pleasure—could reduce the world's vehicles (estimated to be 800,000,000) to a fraction of that number within a few decades. Arth claims that this would be possible if almost all private cars requiring drivers, which are not in use and parked 90% of the time, would be traded for public self-driving taxis that would be in near constant use. This would also allow for getting the appropriate vehicle for the particular need—a bus could come for a group of people, a limousine could come for a special night out, and a Segway could come for a short trip down the street for one person. Children could be chauffeured in supervised safety, DUIs would no longer exist, and 41,000 lives could be saved each year in the U.S. alone.

Key players

International

The European Union has a multi-billion Euro programme to support Research and Development by ad-hoc consortia from the various member countries, called Framework Programmes for Research and Technological Development. Several of these projects pertain to the subject of driverless cars, e.g.:

INRIA's La Route Automatisée project gathered much useful data about the actual and possible deployments of Driverless Cars for public transport. The main system discussed is based on FROG.

Many of the EU-sponsored projects are coordinated by a group called Ertico.

There are several national associations around the world that are active in research in the field of intelligent transportation systems, a term that seems to encompass anything which applies technology to the improvement of transport. In recent years there has been a trend in this field to move efforts away from the more visionary projects, such as driverless cars, to the more short-term, such as public transport and traffic management. Many of these organizations are government sponsored, and they all cooperate at some level or another. Some of the countries involved are: USA, IEEE ITS Society, Australia, South Korea, Taiwan, India--(specifically Intelligent vehicles), and Japan, specifically a cruise assist effort (see below). A more complete list of its organizations can be found here.

Governments

USA:

ITS - Turner-Fairbank Highway Research Center

Ice Detection and Cooperative Curve Warning / Current AVCS Deployment - NTL Catalog

Universities and professional bodies

UC Berkeley - California PATH

MIT Media Lab CityCar

VisLab: Artificial Vision and Intelligent Systems Lab at University of Parma, Italy

Virginia Tech

Austin Robot Technology / UT Austin

IEEE has a Society (the Intelligent Transportation Systems Society), runs an important scientific Journal, and organizes conferences

Japanese Automobile Research Institution

Advanced Cruise-Assist Highway System Research Organization

Carnegie Mellon University Navlab

GrayMatter Inc. - a division of the Gray Team.

Institute of Autonomous Systems Technology: at Bundeswehr University of Munich

Private companies

Google driverless car

General Motors EN-V

Voluntary and hobbyist groups

Autonomous Robots Magazine

American Industrial Magic http://aimagic.org entered 3 vehicles in the 2004 DARPA challenge.

Open Source Driverless Car Project (Python/C++) http://bitbucket.org/djlyon/smp-driverless-car-robot

In film

KITT, the automated Pontiac TransAm in the TV series Knight Rider could drive by itself upon command

The 1989 film Batman, starring Michael Keaton, the Batmobile is shown to be able to drive itself to Batman's current location.

The 1990 film Total Recall, starring Arnold Schwarzenegger, features taxis apparently controlled by artificial intelligence; it is not clear, however, whether these are truly autonomous vehicles or simply conventional vehicles driven by androids.

The 1993 film Demolition Man, starring Sylvester Stallone, set in 2032, features vehicles that can be self-driven or commanded to "Auto Mode" where a voice controlled computer operates the vehicle.

The 1994 film Timecop, starring Jean-Claude Van Damme, set in 2004 and 1994, has cars that can either be self-driven or commanded to drive to specific locations such as "home".

Another Arnold Schwarzenegger movie, The 6th Day (2000), features a driverless car in which Michael Rapaport sets the destination and vehicle drives itself while Rapaport and Schwarzenegger converse.

The 2002 film Minority Report, set in Washington, D.C. in 2054, features an extended chase sequence involving driverless personal cars. The vehicle of protagonist John Anderton is transporting him when its systems are overridden by police in an attempt to bring him into custody.

The 2004 film I, Robot features vehicles with automated driving on future highways, allowing the car to travel safer at higher speeds than if manually controlled. An interesting concept of automated driving in this film is that people aren't trusted to drive manually, as opposed by people not trusting automated driving nowadays.

Anthropomorphic cars (capable of thinking and moving around on their own) have also shown up in movies, such as the series concerning Herbie and the movie Cars. (The name of Volkswagen's 53+1 car was a nod to Herbie;Herbie was conspicuously decorated with the number 53.)

(source:wikipedia)

Autopilot

Autopilot panel of an older Boeing 747 aircraft,.

An autopilot is a mechanical, electrical, or hydraulic system used to guide a vehicle without assistance from a human being. An autopilot can refer specifically to aircraft, self-steering gear for boats, or auto guidance of space craft and missiles. The autopilot of an aircraft is sometimes referred to as "George".

First autopilots

In the early days of aviation, aircraft required the continuous attention of a pilot in order to fly safely. As aircraft range increased allowing flights of many hours, the constant attention led to serious fatigue. An autopilot is designed to perform some of the tasks of the pilot.

The first aircraft autopilot was developed by Sperry Corporation in 1912. The autopilot connected a gyroscopic Heading indicator and attitude indicator to hydraulically operated elevators and rudder (ailerons were not connected as wing dihedral was counted upon to produce the necessary roll stability.) It permitted the aircraft to fly straight and level on a compass course without a pilot's attention, greatly reducing the pilot's workload.

Lawrence Sperry (the son of famous inventor Elmer Sperry) demonstrated it two years later in 1914 at an aviation safety contest held in Paris. At the contest, Lawrence Sperry demonstrated the credibility of the invention were shown by flying the aircraft with his hands away from the controls and visible to onlookers of the contest. This autopilot system was also capable of performing take-off and landing, and the French military command showed immediate interest in the autopilot system. Wiley Post used a Sperry autopilot system to fly alone around the world in less than eight days in 1933.

Further development of the autopilot were performed, such as improved control algorithms and hydraulic servomechanisms. Also, inclusion of additional instrumentation such as the radio-navigation aids made it possible to fly during night and in bad weather. In 1947 a US Air Force C-53 made a transatlantic flight, including takeoff and landing, completely under the control of an autopilot.

In the early 1920s, the Standard Oil tanker J.A Moffet became the first ship to use an autopilot.

Modern autopilots

Not all of the passenger aircraft flying today have an autopilot system. Older and smaller general aviation aircraft especially are still hand-flown, while small airliners with fewer than twenty seats may also be without an autopilot as they are used on short-duration flights with two pilots. The installation of autopilots in aircraft with more than twenty seats is generally made mandatory by international aviation regulations. There are three levels of control in autopilots for smaller aircraft. A single-axis autopilot controls an aircraft in the roll axis only; such autopilots are also known colloquially as "wing levellers", reflecting their limitations. A two-axis autopilot controls an aircraft in the pitch axis as well as roll, and may be little more than a "wing leveller" with limited pitch-oscillation-correcting ability; or it may receive inputs from on-board radio navigation systems to provide true automatic flight guidance once the aircraft has taken off until shortly before landing; or its capabilities may lie somewhere between these two extremes. A three-axis autopilot adds control in the yaw axis and is not required in many small aircraft.

Autopilots in modern complex aircraft are three-axis and generally divide a flight into taxi, takeoff, ascent, level, descent, approach and landing phases. Autopilots exist that automate all of these flight phases except the taxiing. An autopilot-controlled landing on a runway and controlling the aircraft on rollout (i.e. keeping it on the centre of the runway) is known as a CAT IIIb landing or Autoland, available on many major airports' runways today, especially at airports subject to adverse weather phenomena such as fog. Landing, rollout and taxi control to the aircraft parking position is known as CAT IIIc. This is not used to date but may be used in the future. An autopilot is often an integral component of a Flight Management System.

Modern autopilots use computer software to control the aircraft. The software reads the aircraft's current position, and controls a Flight Control System to guide the aircraft. In such a system, besides classic flight controls, many autopilots incorporate thrust control capabilities that can control throttles to optimize the air-speed, and move fuel to different tanks to balance the aircraft in an optimal attitude in the air. Although autopilots handle new or dangerous situations inflexibly, they generally fly an aircraft with a lower fuel-consumption than a human pilot.

The autopilot in a modern large aircraft typically reads its position and the aircraft's attitude from an inertial guidance system. Inertial guidance systems accumulate errors over time. They will incorporate error reduction systems such as the carousel system that rotates once a minute so that any errors are dissipated in different directions and have an overall nulling effect. Error in gyroscopes is known as drift. This is due to physical properties within the system, be it mechanical or laser guided, that corrupt positional data. The disagreements between the two are resolved with digital signal processing, most often a six-dimensional Kalman filter. The six dimensions are usually roll, pitch, yaw, altitude, latitude and longitude. Aircraft may fly routes that have a required performance factor, therefore the amount of error or actual performance factor must be monitored in order to fly those particular routes. The longer the flight the more error accumulates within the system. Radio aids such as DME, DME updates and GPS may be used to correct the aircraft position.

Computer system details

The hardware of an autopilot varies from implementation to implementation, but is generally designed with redundancy and reliability as foremost considerations. For example, the Rockwell Collins AFDS-770 Autopilot Flight Director System used on the Boeing 777, uses triplicated FCP-2002 microprocessors which have been formally verified and are fabricated in a radiation resistant process.

Software and hardware in an autopilot is tightly controlled, and extensive test procedures are put in place.

Some autopilots also use design diversity. In this safety feature, critical software processes will not only run on separate computers and possibly even using different architectures, but each computer will run software created by different engineering teams, often being programmed in different programming languages. It is generally considered unlikely that different engineering teams will make the same mistakes. As the software becomes more expensive and complex, design diversity is becoming less common because fewer engineering companies can afford it. The flight control computers on the Space Shuttle uses this design: there are five computers, four of which redundantly run identical software, and a fifth backup running software that was developed independently. The software on the fifth system provides only the basic functions needed to fly the Shuttle, further reducing any possible commonality with the software running on the four primary systems.

Categories

Instrument-aided landings are defined in categories by the International Civil Aviation Organization. These are dependent upon the required visibility level and the degree to which the landing can be conducted automatically without input by the pilot.

CAT I - This category permits pilots to land with a decision height of 200 ft (61 m) and a forward visibility or Runway Visual Range (RVR) of 550 m. Simplex autopilots are sufficient.

CAT II - This category permits pilots to land with a decision height between 200 ft and 100 ft (≈ 30 m) and a RVR of 300 m. Autopilots have a fail passive requirement.

CAT IIIa -This category permits pilots to land with a decision height as low as 50 ft (15 m) and a RVR of 200 m. It needs a fail-passive autopilot. There must be only a 10−6 probability of landing outside the prescribed area.

CAT IIIb - As IIIa but with the addition of automatic roll out after touchdown incorporated with the pilot taking control some distance along the runway. This category permits pilots to land with a decision height less than 50 feet or no decision height and a forward visibility of 250 ft (76 m, compare this to aircraft size, some of which are now over 70 m long) or 300 ft (91 m) in the United States. For a landing-without-decision aid, a fail-operational autopilot is needed. For this category some form of runway guidance system is needed: at least fail-passive but it needs to be fail-operational for landing without decision height or for RVR below 100 m.

CAT IIIc - As IIIb but without decision height or visibility minimums, also known as "zero-zero".

Fail-passive autopilot: in case of failure, the aircraft stays in a controllable position and the pilot can take control of it to go around or finish landing. It is usually a dual-channel system.

Fail-operational autopilot: in case of a failure below alert height, the approach, flare and landing can still be completed automatically. It is usually a triple-channel system or dual-dual system.

Radio-controlled models

In radio-controlled modelling, and especially RC aircraft and helicopters, an autopilot is usually a set of extra hardware and software that deals with pre-programming the model's flight.

See also

Gyrocompass

Driverless car

(source:wikipedia)

Motion planning

Real-Time Scalable Motion Planning for Crowds

Motion planning (a.k.a., the "navigation problem", the "piano mover's problem") is a term used in robotics for the process of detailing a task into discrete motions.

For example, consider navigating a mobile robot inside a building to a distant waypoint. It should execute this task while avoiding walls and not falling down stairs. A motion planning algorithm would take a description of these tasks as input, and produce the speed and turning commands sent to the robot's wheels. Motion planning algorithms might address robots with a larger number of joints (e.g., industrial manipulators), more complex tasks (e.g. manipulation of objects), different constraints (e.g., a car that can only drive forward), and uncertainty (e.g. imperfect models of the environment or robot).

Motion planning has several robotics applications, such as autonomy, automation, and robot design in CAD software, as well as applications in other fields, such as animating digital characters, architectural design, robotic surgery, and the study of biological molecules.

Concepts

Example of a workspace.

Configuration space of a point-sized robot. White = Cfree, gray = Cobs.

Configuration space for a rectangular translating robot (pictured red). White =Cfree, gray = Cobs, where dark gray = the objects, light gray = configurations where the robot would touch an object or leave the workspace.

Example of a valid path.

Example of an invalid path.

Example of a road map.

A basic motion planning problem is to produce a continuous motion that connects a start configuration S and a goal configuration G, while avoiding collision with known obstacles. The robot and obstacle geometry is described in a 2D or 3D workspace, while the motion is represented as a path in (possibly higher-dimensional) configuration space.

Configuration Space

A configuration describes the pose of the robot, and the configuration space C is the set of all possible configurations. For example:

If the robot is a single point (zero-sized) translating in a 2-dimensional plane (the workspace), C is a plane, and a configuration can be represented using two parameters (x, y).

If the robot is a 2D shape that can translate and rotate, the workspace is still 2-dimensional. However, C is the special Euclidean group SE(2) = R2 SO(2) (where SO(2) is the special orthogonal group of 2D rotations), and a configuration can be represented using 3 parameters (x, y, θ).

If the robot is solid 3D shape that can translate and rotate, the workspace is 3-dimensional, but C is the special Euclidean group SE(3) = R3 SO(3), and a configuration requires 6 parameters: (x, y, z) for translation, and Euler angles (α, β, γ).

If the robot is a fixed-base manipulator with N revolute joints (and no closed-loops), C is N-dimensional.

Free Space

The set of configurations that avoids collision with obstacles is called the free space Cfree. The complement of Cfree in C is called the obstacle or forbidden region.

Often, it is prohibitively difficult to explicitly compute the shape of Cfree. However, testing whether a given configuration is in Cfree is efficient. First, forward kinematics determine the position of the robot's geometry, and collision detection tests if the robot's geometry collides with the environment's geometry.

Algorithms

Low-dimensional problems can be solved with grid-based algorithms that overlay a grid on top of configuration space, or geometric algorithms that compute the shape and connectivity of Cfree.

Exact motion planning for high-dimensional systems under complex constraints is computationally intractable. Potential-field algorithms are efficient, but fall prey to local minima (an exception is the harmonic potential fields). Sampling-based algorithms avoid the problem of local minima, and solve many problems quite quickly. They are unable to determine that no path exists, but they have a probability of failure that decreases to zero as more time is spent.

Sampling-based algorithms are currently considered state-of-the-art for motion planning in high-dimensional spaces, and have been applied to problems which have dozens or even hundreds of dimensions (robotic manipulators, biological molecules, animated digital characters, and legged robots).

Grid-Based Search

Grid-based approaches overlay a grid on configuration space, and assume each configuration is identified with a grid point. At each grid point, the robot is allowed to move to adjacent grid points as long as the line between them is completely contained within Cfree (this is tested with collision detection). This discretizes the set of actions, and search algorithms (like A*) are used to find a path from the start to the goal.

These approaches require setting a grid resolution. Search is faster with coarser grids, but the algorithm will fail to find paths through narrow portions of Cfree. Furthermore, the number of points on the grid grows exponentially in the configuration space dimension, which make them inappropriate for high-dimensional problems.

Traditional grid-based approaches produce paths whose heading changes are constrained to multiples of a given base angle, often resulting in suboptimal paths. Any-angle path planning approaches find shorter paths by propagating information along grid edges (to search fast) without constraining their paths to grid edges (to find short paths).

Grid-based approaches often need to search repeatedly, for example, when the knowledge of the robot about the configuration space changes or the configuration space itself changes during path following. Incremental heuristic search algorithms replan fast by using experience with the previous similar path-planning problems to speed up their search for the current one.

Geometric Algorithms

Point robots among polygonal obstacles

Visibility graph

Cell decomposition

Translating objects among obstacles

Minkowski sum

Potential Fields

One approach is to treat the robot's configuration as a point in a potential field that combines attraction to the goal, and repulsion from obstacles. The resulting trajectory is output as the path. This approach has advantages in that the trajectory is produced with little computation. However, they can become trapped in local minima of the potential field, and fail to find a path.

Sampling-Based Algorithms

Sampling-based algorithms represent the configuration space with a roadmap of sampled configurations. A basic algorithm samples N configurations in C, and retains those in Cfree to use as milestones. A roadmap is then constructed that connects two milestones P and Q if the line segment PQ is completely in Cfree. Again, collision detection is used to test inclusion in Cfree. To find a path that connects S and G, they are added to the roadmap. If a path in the roadmap links S and G , the planner succeeds, and returns that path. If not, the reason is not definitive: either there is no path in Cfree, or the planner did not sample enough milestones.

These algorithms work well for high-dimensional configuration spaces, because unlike combinatorial algorithms, their running time is not (explicitly) exponentially dependent on the dimension of C. They are also (generally) substantially easier to implement. They are probabilistically complete, meaning the probability that they will produce a solution approaches 1 as more time is spent. However, they cannot determine if no solution exists.

Given basic visibility conditions on Cfree, it has been proven that as the number of configurations N grows higher, the probability that the above algorithm finds a solution approaches 1 exponentially . Visibility is not explicitly dependent on the dimension of C; it is possible to have a high-dimensional space with "good" visibility or a low dimensional space with "poor" visibility. The experimental success of sample-based methods suggests that most commonly seen spaces have good visibility.

There are many variants of this basic scheme:

It is typically much faster to only test segments between nearby pairs of milestones, rather than all pairs.

Nonuniform sampling distributions attempt to place more milestones in areas that improve the connectivity of the roadmap.

Quasirandom samples typically produce a better covering of configuration space than pseudorandom ones, though some recent work argues that the effect of the source of randomness is minimal compared to the effect of the sampling distribution.

If only one or a few planning queries are needed, it is not always necessary to construct a roadmap of the entire space. Tree-growing variants are typically faster for this case (single-query planning). Roadmaps are still useful if many queries are to be made on the same space (multi-query planning)

Completeness and Performance

A motion planner is said to be complete if the planner always produces a feasible path, when one exists. Most complete algorithms are geometry-based. The performance of a complete planner is assessed by its computational complexity.

Resolution completeness is the property that the planner is guaranteed to find a path if the resolution of an underlying grid is fine enough. Most resolution complete planners are grid-based. The computational complexity of resolution complete planners is dependent on the number of points in the underlying grid, which is O(1/hd), where h is the resolution (the length of one side of a grid cell) and d is the configuration space dimension.

Probabilistic completeness is the property that as more “work” is performed, the probability that the planner fails to find a path, if one exists, asymptotically approaches zero. Several sample-based methods are probabilistically complete. The performance of a probabilistically complete planner is measured by the rate of convergence.

Incomplete planners do not always produce a feasible path when one exists. Sometimes incomplete planners do work well in practice.

Problem Variants

Many algorithms have been developed to handle variants of this basic problem.

Differential Constraints

Holonomic

Manipulator arms (with dynamics)

Nonholonomic

Cars

Unicycles

Planes

Acceleration bounded systems

Moving obstacles (time cannot go backward)

Bevel-tip steerable needle

Differential Drive Robots

Optimality Constraints

Hybrid Systems

Hybrid systems are those that mix discrete and continuous behavior. Examples of such systems are:

Robotic manipulation

Mechanical assembly

Legged robot locomotion

Reconfigurable robots

Uncertainty

Motion uncertainty

Missing information

Active sensing

Sensorless planning

Applications

Robot navigation

Automation

The driverless car

Robotic surgery

Digital character animation

protein folding

Safety and accessibility in computer-aided architectural design

(source:wikipedia)