![](bg6.png)
0.08
0.1
0.12
0.14
0.16
0.18
0.2
0.22
0 50 100 150 200
error per meter (m)
time (s)
rolling average w/o learning
average w/o learning
rolling average with learning
average with learning
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45
0
1
2
3
4
5
6
7
8
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45
0
1
2
3
4
5
6
7
8
Figure 6: Experimental results: Left: The rolling and total average of the two setups. The learning variant outperforms the
static version. Middle and right: Histogram showing the absolute error distribution without and with learning. Right: Large
errors occur less frequent.
7 SUMMARY AND
CONCLUSIONS
In this paper we presented a navigation system based
on predefined motion templates which are combined
with a tree-search technique to achieve efficient tra-
jectories. We introduced Velocity Grids to represent
difficult or impassable terrain by means of a maxi-
mum admissible velocity. The system has the ability
to adjust the motion templates according to the actual
robot movement, which is measured by an IMU, GPS,
and lidar-based motion estimation. The soundness of
our approach has been shown both in real-word navi-
gation tasks. The system proved that it can navigate a
robot through an obstacle course, and that it is able to
adapt to different surfaces quickly.
Future work will focus on improving the perfor-
mance of the motion learning and adapting mecha-
nisms. The current countermeasure against pattern
pool depletion causes temporarily outdated predic-
tions. The most promising remedy would be to prop-
agate the changes of one Motion Pattern to all others,
provided that a sufficiently precise online approxima-
tion can be found. Another improvement worth inves-
tigating is the usage of the pattern deviation that is de-
termined by the learning module as uncertainty mea-
sure in order to optimize the Motion Patterns’ safety
margins.
REFERENCES
Brock, O. and Khatib., O. (1999). High-speed navi-
gation using the global dynamic window approach.
In Proc. of the IEEE International Conference on
Robotics & Automation (ICRA).
Bruch, M., Gilbreath, G., and Muelhauser, J. (2002). Ac-
curate waypoint navigation using non-differential gps.
In Proc. of AUVSI Unmanned Systems.
Brunner, M., Schulz, D., and Cremers, A. B. (2010). Po-
sition estimation of mobile robots considering char-
acteristic terrain model. In submitted to the Interna-
tional Conference on Informatics in Control, Automa-
tion and Robotics (ICINCO).
Fox, D., Burgard, W., and Thrun, S. (1997). The dy-
namic window approach to collision avoidance. IEEE
Robotics & Automation Magazine, 4(1):23–33.
Ko, J., Klein, D., Fox, D., and Haehnel, D. (2002). Vision-
based Monte-Carlo self-localization for a mobile ser-
vice robot acting as shopping assistant in a home store.
In Proc. of the IEEE/RSJ International Conference on
Intelligent Robots and Systems (IROS).
Neumann, G., Maass, W., and Peters, J. (2009). Learning
complex motions by sequencing simpler motion tem-
plates. In Proc. of the International Conference on
Machine Learning (ICML).
Pivtoraiko, M., Knepper, R. A., and Kelly, A. (2009). Dif-
ferentially constrained mobile robot motion planning
in state lattices. J. Field Robot., 26(3):308–333.
Seyr, M., Jakubek, S., Novak, G., and Dellaert, F. (2005).
Neural network predictive trajectory tracking of an
autonomous two-wheeled mobile robot. In Interna-
tional Federation of Automatic Control (IFAC) World
Congress.
Thrun, S., Montemerlo, M., Dahlkamp, H., Stavens, D.,
Aron, A., Diebel, J., Fong, P., Gale, J., Halpenny,
M., Hoffmann, G., Lau, K., Oakley, C., Palatucci, M.,
Pratt, V., Stang, P., and Strohband, S. (2006). Winning
the darpa grand challenge. Journal of Field Robotics.
V. Hundelshausen, F., Himmelsbach, M., Hecker, F.,
Mueller, A., and Wuensche, H.-J. (2008). Driving
with tentacles: Integral structures for sensing and mo-
tion. J. Field Robot., 25(9):640–673.
Vallespi, C. and Stentz, A. (2008). Prior data and ker-
nel conditional random fields for obstacle detection.
In Proceedings of the Robotics: Science and Systems
(RSS).
Wellington, C., Courville, A., and Stentz, A. (2005). Inter-
acting markov random fields for simultaneous terrain
modeling and obstacle detection. In Proceedings of
the Robotics: Science and Systems (RSS).
OFFROAD NAVIGATION USING ADAPTABLE MOTION PATTERNS
191