Odometric navigation with matching of landscape features

This paper presents a technique to localize and guide outdoors vehicles by both vehicle odometry and landmark localization. In this case, we use reflective poles spaced out about 50 m apart and located at the front of the vehicle. The LADAR detects distances, angles, speed, and the position of obstacles within 20 m. Odometry is performed by a steering sensor and encoder attached at the vehicle differential. Using this information, the embedded system matches the vehicle's computed position with the location of the poles. Finally, passenger safety is ensured by an anti-collision subsystem thanks to the LADAR sensor.


INTRODUCTION
Most European cities face numerous challenges associated with the use of private vehicles.Problems include road congestion, energy expenditure, noise and pollution, all of which degrade the quality of urban life.Historical city centres are facing severe problems: traditional commerce declines, moves to the periphery, and they become less attractive to tourists.

PROJECTBACKGROUTiD
The CTS (Cybemetic Transportation System) concept started with car-sharing: a fleet of individual vehicles shared among a relatmely large number of users, offers the possibility of using a car for some time or having a car available at both ends of a train trip.These systems are increasingly popular in Switzerland and Germany.They work well in specific areas where the demand is properly structured, but have yet to offer a door to door service: currently, the vehicles are only available at a few locations and have to be returned.Modem fleet management technologies have recently improved the service, e.g.allowing short trips without the need to return the car to its point of origin.Such systems are called station-car systems and have been developed since the mid-90's.For example, F'raxitL-le and Liselec Projects in France, City-Car in Switzerland, IntelliShare in the USA, Crayon in Japan (Fig. l), just to name a few.Specific vehicles, well adapted to city driving, have generally been used for these systems: small size, convenient, energy efficient, quiet, often based on electric power.They even compete with public transportation in tams of energy consumption on per passenger-km basis.However, these systems have not generally proven that they can compete economically, one reason being limited vehicles availability in too few locations, thus limiting the number of potential customers (see European Project Utopia).
Cybercars is a car-sharing concept that can dramatically improve upon exist fleet management systems.Here, vehicles move by themselves in response to the demand of passengers or the need to move goods.

WHAT IS A CYBERCAR?
Cybercars are road vehicles (Fig. 2) with fully automated driving capabilities.A fleet of such vehicles forms a transportation system (CTS), for passengers or goods, on a network of roads with on-demand and door-to-door capability.The fleet of cars is under control of a central management system in order to meet particular demands in a particular environment.At the initial stages, cybercars are designed for short trips at low speed in an urban environment or in private grounds.
In the long term, cybercars could also run autonomously at high speed on dedicated tracks.With the development of the CTS infhstructures, private cars with fully autonomous driving capabilities could also be allowed on these infrastructures while maintaining their manual mode on standard roads.First, they provide reduction of congestion, and better traffic flow, air quality and energy conservation.
Second, in automatic mode, the systems are very safe.In fact, there is no need for a driver's license, so anyone can use it, including the elderly and person with handicaps.Third, the cars can be moved easily .fromone location to another, using fully autonomous driving platoon formations with a single driver.Fourth, the cars can drive autonomously to a remote parking area when not needed, hence leaving valuable urban space free for pedestrians and cyclists.Fifth, the concept and technologies are also appropriate for delivery of goods in city centers and even for garbage collection: the same infixstructure could be used by specifically adapted vehicles with delivery (or collection) "boxes".Finally, flexible design will make it possible to optimize the overall system performance, taking into account the needs and requirements of the private consumer, the system operator and the public (e.g.municipality), permitting the system to operate in different modes at different times of the day, week and year.
Several companies and research organizations have been involved, during the last years, in the development of these new vehicles (Fig. 3).Most of the actors have established a partnership to launch the CyberCars project.

AUTOMATED GUIDED VEHICLE
The testing vehicle is an outdoor electrical car made by Yamaha based on a golf car frame, equipped with sensors and an embedded PC.Three driving modes are available: manual, wire-guided and automatic.For this experiment, we use the vehicle in the automatic mode i.e. speed and steering controls are managed by our algorithms.The following are brief descriptions of the sensors:

Encoder
One low-cost encoder in quadrature is attached to the differential of the vehicle.It directly measures mean rotation of the two rear wheels.

Steering sensor
One potentiometer is attached to the steering shaft.

Gyrometer
speed at the frequency of 30 Hz.

GPS
One low-cost GPS receiver [2] provides the geographical coordinates at the frequency of 1 Hz.

m h ? R
The LADAR, a laser scanner [3], transmits a pulsed laser beam which uses a rotating mirror to scan an area in front of the vehicle.Within this area of 20 m x 20 m, distances and angles to poles are measured such that position and speed are available to the vehicle computer via a CAN bus at the kquency of 10 Hz.
Xt represents the perturbation on the last estimated state Xo,.
The linearization of ( 2) by a first order Taylor series approximation makes usable the standard Kalman formalism.
Previous equation forms the Kalman equation of model.The state noise is 6, represented by centered and independent white noise with known variance.
"If the problem is sufficiently observable, then the deviations between the estimated trajectory and the actual trajectory will remain sufficiently small that the linearization assumption is valid [5][6J" [71.

GPS observations
The GPS provides dead-reckoning feedback by measuring the vehicle position, designated as xgp, ym and \vW, at the frequency of 1 Hz: The measure noise is E, represented by centered and independent white noise with known variance.

LADAR observationr
As stated earlier, the LALIAR provides dead-reckoning feedback by measuring the vehicle location with respect to the reflective poles spaced throughout the vehicle course.
Poles are easily identified by the LADup to a range of 12 m away at the frequency of 10 Hz.The LADAR detects the nearest pole (point A, Fig. 4) and tracks its positions until the vehicle passes the pole.The measurement provides the position and orientation of pole in the vehicle frame, here designated by X,, Y,, and "A.
Secondly, the position and orientation of pole in the global &me is known, designated as XA, YA and \ v ~.The vehicle position is given by the formula: The measure noise is y.represented by centered and independent white noise with known variance.The vehicle updates its position if a pole is detected by the LADAR and identified in the GIs database (Geographical Information System) of vehicle.

VEHICLE GUIDANCE
A central management system sends the path to follow, a list of positions and orientations, to the vehicle via a Wave-LAN network.Then, the vehicle guidance is ensured by the target planning and the trajectory following subsystems (Fig. 5) as describe below.Obstacle positions are detected by the LADAR and projected onto the global coordinate system.Based upon this data and the specified minimum distance of collision, vehicle speed is reduced until the obstacle has cleared the path.

Trajectory following 5.2.1. High level
Driving commands are computed depending on the distance between the vehicle position and the target position.The error, designated by e,, ey and e,+,, is equal to the coordinates of target position in the vehicle f h n e : Then, we apply the following law of command modified from [8]: Here, vc and cp, represent the commands of speed and steering angle.v, , is the maximum speed of the vehicle associated with the target position.Kxy and I & are constant parameters tuned to obtain an optimal trajectory following.

Low level
PI controllers are used to control the speed and the steering angle of vehicle.At kh time iteration, cps represents the control of steering angle updated at the frequency of 30 Hz. Kp and KI are proportional and integral gains tuned to obtain the lateral control of vehicle.

IMPLEMENTATION
To guarantee safe and reliable performance, we have implemented our guided transportation system with RTAI Linux 191.Besides the well-known benefits of strict timing constraints, RTAI also provides numerous features that aid our development.In our case, we make extensive use of operating modules, communicating asynchronously with mailboxes to achieve the tasks of detection, navigation, and controls.This is schematically shown Fig.

CONCLUSION
In conclusion, we have presented a navigation technique for outdoor low-speed vehicles which improves the odometric navigation with matching of landscape features.Software architecture

Fig. :
Fig.: Frog cybercar in Schiphol The advantages of autonomous driving capabilities and the new transportation systems, based on environmental fiiendly vehicles, are numerous.

5. 1 .
Target planning A target position, designated by x,, yt and \vt in the global frame, is computed at the frequency of 10 Hz depending on the path and the nearest obstacle.
Fip.: Experimental result by irajectoxy following {circles} with matching of poles {crosses)