Ground planeScanning plane
4: for every Particle do
6.2. Future Work
Several aspects of this work can be further explored. One aspect is the possible release of some assumptions made by adding additional sensors.
Relieve of Assumptions. The known-height of the LIDAR assumption could be dropped given an Inertial Measurement Unit (IMU) is added to the robot. If the IMU could sense
6.2 FUTURE WORK
the LIDARs orientations accurately, the sensors height for each scan could be estimated by using geometrical constraints given flat ground that is visible in the scan.
Taking the robot from indoor environments to urban environments might uncover new interesting challenges. When leaving buildings it might be necessary to drop the as-sumption of flat ground. Given an gradually built elevation map, fixed height of the laser rangefinder and an IMU the robots pose in 3D could still be estimated even it is not oper-ating on flat ground. Since the estimation of the ground height is likely to be only locally accurate, prior knowledge like a large-scale street map with elevation information could be used to correct accumulated errors.
Loop Closing. Currently the work introduced is not able to detect and close loops.
This is one important building block to enable reliable large scale SLAM. A Graph-SLAM based approach could be implemented. A future loop detection mechanism might work better than the normal 2D case due the richer 3D information available.
Multiple Hypotheses. The sampling based approach introduced in Section 4.3.4 can be used to keep track of multiple hypotheses in the matching stage. Ambiguities in the environment can lead to several results, each of them providing a good matching result.
Treating all these results as valid hypotheses allows to use them in offline optimization method. Assuming a locally smooth trajectory (temporal as well as spatial) could allow to pick the correct hypotheses in an offline stage. Further prior information about the environment could be applied to pick the hypotheses that models the environment best.
Optimization using prior knowledge. This thesis makes already use of prior knowl-edge about the environment when assuming the ground is flat and that there are vertical structures. The 3D point clouds generated show that the representation of walls is locally accurate, but small errors accumulate leading to slightly curved or inclined walls. Walls in buildings are almost exclusively vertical planes - therefore optimization methods could use this information to correct localization errors - and could reduce the expensive point-representation into a point-representation by planes.
46
BIBLIOGRAPHY
Besl, P. & McKay, H. A method for registration of 3-d shapes. IEEE Transactions on Pattern Analysis and Machine Intelligenc.
Burgard, W., Fox, D., Jans, H., Matenar, C., & Thrun, S. (1999). Sonar-based mapping of large-scale mobile robot environments using em. International Conference on Machine Learning.
Cheeseman, P. & Smith, R. (1986). On the representation and estimation of spatial uncer-tainty. International Journal of Robotics Research.
Derenick, J., Miller, T., Spletzer, J., Kushleyev, A., Foote, T., Stewart, J., Bohren, A., & Lee, D.
(2009). The sick lidar matlab/c++ toolbox: Software for rapidly interfacing/configuring sick lidars with applications to real-time experimental robotics. IEEE/RSJ International Conference on Intelligent Robots and Systems.
Elfes, A. (1987). Sonar-based real-world mapping and navigation. IEEE Transactions on Robotics and Automation.
Frueh, C. & Zakhor, A. (2004). An automated method for large-scale, ground-based city model aquisition. International Journal on Computer Vision.
Harrison, A. & Newman, P. (2008). High quality 3d laser ranging under general vehicle motion. IEEE International Conference on Robotics and Automation, 7–12.
Li, G. & Liu, Y. (2007). An algorithm for extrinsic parameters calibration of a camera and a laser range finder using line features. International Conference on Intelligent Robots and Systems.
Lu, F. & Milios, E. (1997). Robot pose estimation in unknown environments by matching 2d range scans. Journal of Intelligent and Robotic Systems, 18(3), 249275.
Martin, M. & Moravec, H. (1996). Robot evidence grids. Technical report, Carnegie Mellon University.
BIBLIOGRAPHY
Nuechter, A., Lingemann, K., Hertzberg, K., & Surman, H. (2005). 6d slam with approxi-mate data association. International Conference on Advanced Robotics.
Pless, R. & Zhang, Q. (2004). Extrinsic calibration of a camera and laser range finder. Inter-national Conference on Intelligent Robots and Systems.
Rusinkiewicz, S. & Levoy, M. (2001). Efficient variants of the icp algorithm. Technical report, Stanford University.
Ryde, J. & Hu, H. (2007). Mobile robot 3d perception and mapping with multi-resolution occupancy lists. IEEE International Conference on Mechatronics and Automation.
Schulz, W. H. (2007). Landslide susceptibility revealed by lidar imagery and historical records, seattle, washington. Engineering Geology, 89(1-2), 67 – 87.
Thrun, S., Burgard, W., & Fox, D. (2000). A real time algorithm for mobile robot map-ping with applications to multi robot and 3d mapmap-ping. IEEE International Conference on Robotics and Automation.
Thrun, S., Burgard, W., & Fox, D. (2005). Probabilistic Robotics. Cambridge, Massachusetts:
The MIT Press.
Thrun, S., Ghahramini, Z., Koller, D., Durrant-Whyte, H., & Ng, A. (2002). Simultaneous mapping and localization with sparse extended information filters. Proceedings of the Fifth International Workshop on Algorithic Foundations of Robotics.
Thrun, S., Haehnel, D., Burgard, W., Ferguson, D., Montemerlo, M., Triebel, R., Baker, C., &
Whittaker, W. (2003). A system for volumetric robotic mining in abandoned mines. IEEE International Conference on Robotics and Automation.
Wulf, O. & Wagner, B. (2003). Fast 3d scanning methods for laser measurement systems.
Technical report, Institute for Systems Engineering, University of Hannover, Germany.
Zhao, H., Chen, Y., & Shibasaki, R. (2007). An efficient extrinsic calibration of a multiple laser scanners and cameras’ sensor system on a mobile platform. IEEE Inteligent Vehicle Symposium.
48
Document Log:
Manuscript Version 5 — 3 August 2009) Typeset by LATEX — 5 August 2009
ANDREASDOPFER
THEROBOTPERCEPTION ANDLEARNINGLAB., DEPARTMENT OFCOMPUTERSCIENCE ANDINFORMA -TIONENGINEERING, NATIONALTAIWANUNIVERSITY, NO.1, SEC. 4, ROOSEVELTROAD, DA-ANDISTRICT, TAIPEICITY, 106, TAIWAN, Tel. : (+886) 2-3366-4888EXT.407
E-mail address: [email protected]