| Home | Research | Interests | Notes | IMS Home |
Humans have an inborn technique of correcting their ideas about their own location. Whenever you go someplace you know only a little bit, there is this sense of trying to recognize familiar signs and once you obtained the right impression, suddently your internal reference frame jumps a known state. Even sometimes, this reference frame is held up for quite a while, only to be replaced by a totally different one, because you got a completely contradicting but even more correct piece of information.
I believe, that a similar strategy could lead to robust positioning independend of the nature of sensors used. The nature and quality of sensors used will only add in a positive way to the overall knowledge, but cannot really constrain it anymore. Moreover the implicit modelling of uncertainty over the current estimate can be well used by applications to provide adaptive user interfaces.
Positioning techniques always work relative to a certain frame of reference which is established by the technology itself. Some reference frames are relative to the user itself such as vision based inside-out tracking. In principle the different reference frames should always build a coherent set without contradictions. In practice, the errors associated with measurements can lead to large contradictions between the reference frames. Statistical methods are typically used to compute a merged estimate of position from different sensors. However, current work typically focuses on a predefined set of sensors and their properties. Then an appropriate model is created to merge the estimates.
It would be interesting to create a framework that is independend to the sensors itself and allow for arbitrary estimates. How could such a framework look like ?
It should support the maintenance of a number of hypotheses at the same time.
It should allow to deduce constraints from the hypotheses that can be used to differentiate and test the hypotheses. This would allow to create a system that actively seeks to improve its estimates.
It should provide a general framework or model to incorporate any kind of location sensor. What would be the appropriate reference frame in this case ? It appears that the only common and usable point of reference is the object to be located itself ! By inverting the measurement relationships we can ground all the hypotheses in one thing, the object itself. This is similar to DesCartes observation, to ground knowledge about existence in oneself :).