244 
7 CONCLUSIONS 
We have studied the task of removing arbitrary objects one by one from a heap without the use of object models. 
Instead we rely on range data from two range sensors which are arranged in such a way as to capture as much 
as possible of the object surfaces. The two data sets are integrated and the result is segmented into connected 
components which are regarded as object hypotheses. We are, however, interested not so much in the objects 
themselves but rather in the identification of grasping opportunities. It has been shown that all knowledge 
necessary for their recognition can be extracted from the range data and from the gripper properties. We have 
found it useful to decompose the identification of grasping opportunities into two steps: with the help of two 
heuristics we first choose a preferred “region of action”. Then, in this restricted domain, we acquire the features 
which are necessary for a good grasp. They can be viewed as evidence for the presence of such an opportunity. 
The accumulated evidence allows the decision to grasp or not. We view the object heap and the gripper in its 
pre-grasp pose in a renderer. In the very near future the actual grasping will be implemented. 
8 ACKNOWLEDGEMENT 
This work was supported by the Swiss National Science foundation, grant NFP/SPP-IF 5003-34415. 
REFERENCES 
[Boissonnat 1982] J.-D. Boissonnat. Stable matching between a hand structure and an object silhouette. IEEE 
Trans. on Pattern Anal. and Machine Intell., 4(6):603-612, November 1982. 
[Ikeuchi and Hebert 1990] K. Ikeuchi and M. Hebert. Task oriented vision. In Image Understanding Workshop, 
pages 497—507, 1990. 
[Mulgaonkar et al. 1992] P.G. Mulgaonkar, C.K. Cowan, and J. DeCurtins. Understanding object configurations 
using range images. IEEE Trans. on Pattern Anal. and Machine Intelligence, 14(2):303-307, 1992. 
[Rutishauser et al. 1994] M. Rutishauser, M. Stricker, and M. Trobina. Merging range images of arbitrarily 
shaped objects. In Proceedings of the Conference on Computer Vision and Pattern Recognition (CVPR'94), 
pages 573-580, 1994. 
[Stansfield 1991] S.A. Stansfield. Robotic Grasping of Unknown Objects: A Knowledge-based Approach. The 
International Journal of Robotics Research, 10(4):314-326, August 1991. 
[Stark et al. 1993] L. Stark, L.O. Hall, and K.W. Bowyer. Methods for combination of evidence in function- 
based 3-d object recognition. International Journal of Pattern Recognition and Artificial Intelligence, 
7(3):573—594, 1993. 
[Trobina and Leonardis 1995] M. Trobina and A. Leonardis. Grasping arbitrarily shaped 3-d objects from a 
pile. To appear in Proceedings IEEE Conference on Robotics and Automation, Nagoya, 1995. 
[Tsikos and Bajcsy 1991] C.J. Tsikos and R.K. Bajcsy. Segmentation via Manipulation. IEEE Transactions on 
Robotics and Automation, 7(3):306-319, June 1991. 
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences", Zurich, March 22-24 1995