algorithms. The matching algorithms themselves are 
part of the block comparison in figure 10. The dif- 
ference signals are weighted according to the actual 
situation and the accuracy of the measurements and 
are used as corrections on the model world. This is 
for example performed within the update steps of the 
Kalman filter discussed in section 6. Thus the model 
world is driven to converge towards the real world. 
The advantage of this model based processing 
scheme stems from the fact that the model world sim- 
ulated on the computer provides considerably more 
detailed information of the ship and its environment 
than the real world sensors themselves. As explained 
in the previous sections, a precise estimate of the own 
ship's position, a detailed reconstruction of the navi- 
gation environment and a dependable observation of 
the actual traffic situation can be obtained from the 
sensor processing level. Therefore all information 
required at the control level is provided. The most 
important task of this level is the trajectory genera- 
tion for the own ship determining the control inputs to 
engine and rudder. 
8 PLANNING AN ACTUAL TRAJECTORY 
Trajectory generation is done in two steps: First, 
an optimal trajectory for upstream and downstream 
travel is computed off-line, assuming the absence of 
foreign ships. However, time invariant environmental 
constraints as well as ship dependent dynamics are 
taken into account. In addition to this, it is possible to 
compute a set of trajectories for different water levels. 
These trajectories are stored in the electronic chart as 
ideal guiding lines. 
For the case that the actual traffic situation results 
in interferences with other ships, a second step of 
on-line trajectory recomputation has to follow in the 
sequel. This second step uses the ideal guiding lines 
and the limits of navigable water as well as traffic 
rules stored in the chart. This a-priori information is 
combined with the results of the multiple-target track- 
ing algorithm. Here the navigator may also interact, 
telling the system how foreign ships are to be passed 
and how they should be encountered. This allows 
to incorporate information resulting from the commu- 
nication with navigators of other ships or from other 
external sources. 
All these sources of information are combined in the 
on-line computation. The algorithm employed starts 
from a coarse grid superimposed over the waterway. 
A risk function is assigned to every point in the grid. 
This function consists of a constant part derived from 
the chart information and a time-varying part repre- 
senting the results of the multiple-target tracking al- 
gorithm. A foreign ship is taken into account at a 
predicted place of encounter. The second step in the 
on-line calculations is a search algorithm resulting in 
possibly several trajectories through the grid and a 
74 
cumulative risk for each trajectory. Finally, one of 
these trajectories is selected as input for the control 
task. 
The control task is implemented as a linear state 
controller designed for variable command control [3]. 
This task generates the signals acting on the rudder 
and engine throttle. For large ships in narrow canals 
this controller is not sufficient. Current research fo- 
cuses on the development of new control concepts 
incorporating results from nonlinear and predictive 
control theory. 
References 
[1] Blackman, S. S.: Multiple-Target Tracking with 
Radar Applications. Artech House, Norwood, MA, 
1986. 
[2] Gelb, A. (Hrsg.): Applied Optimal Estimation. The 
M.I.T. Press, Cambridge, Massachusetts, 1974. 
[3] Gilles, E. D., Neul, R., Plocher, T., und Kabatek, 
U.: Ein integriertes Navigationssystem für Bin- 
nenschiffe. Automatisierungstechnik 38 (1990), 
S. 202-209, 247—257. 
[4] Kabatek, U., Sandler, M., Neul, H., und Gilles, 
E. D.: Eine elektronische FluBkarte als Wissens- 
basis in einem integrierten Navigationssystem. 
Zeitschrift für Vermessungswesen 117 (1992), S. 
35-45. 
[b] Neul, R.: Positionsbestimmung eines navigieren- 
den Schiffes durch kartengestützte Radarbildver- 
arbeitung, volume 323 of Fortschittsberichte VDI 
Reihe 8. VDI-Verlag, Düsseldorf, 1993. 
[6] Plocher, T.: Einsatz von Kalman-Filtern und 
Bayesschen Schátzverfahren zur Verfolgung be- 
wegter Objekte in Bildsequenzen, volume 320 
of Fortschittsberichte VDI Reihe 8. VDI-Verlag, 
Düsseldorf, 1993. 
[7] Plocher, T. und Gilles, E. D.: Rekursive Objek- 
tverfolgung in Bildsequenzen. Automatisierung- 
stechnik 40 (1992), S. 14—20,59-63. 
[8] Reid, D.: An algorithm for tracking multiple tar- 
gets. IEEE Trans. Automat. Contr. 24(12) (1979), 
S. 843-854. 
Cet a 
serva 
navig 
emplo 
laser 
combi 
élect 
navir 
navir 
moind 
capte 
élect 
mesur 
d'un 
cible 
et év 
du ss 
traje 
métho 
La Xst 
navig 
circu