Domain-specific program synthesis can automatically generate high quality code in complex domains from succinct specifications, but the range of programs which can be generated by a given synthesis system is typically narrow. Obtaining code which falls outside this narrow scope necessitates either 1) extension of the code generator, which is usually very expensive, or 2) manual modification of the generated code, which is often difficult and which must be redone whenever changes are made to the program specification. In this paper, we describe adaptations and extensions of the AUTOFILTER Kalman filter synthesis system which greatly extend the range of programs which can be generated. Users augment the input specification with a specification of code fragments and how those fragments should interleave with or replace parts of the synthesized filter. This allows users to generate a much wider range of programs without their needing to modify the synthesis system or edit generated code. We demonstrate the usefulness of the approach by applying it to the synthesis of a complex state estimator which combines code from several Kalman filters with user-specified code. The work described in this paper allows the complex design decisions necessary for real-world applications to be reflected in the synthesized code. When executed on simulated input data, the generated state estimator was found to produce comparable estimates to those produced by a hand-coded estimator.
Flexible generation of Kalman filter code
2006 IEEE Aerospace Conference ; 8 pp.
2006-01-01
9318615 byte
Conference paper
Electronic Resource
English
Rapid Exploration of the Design Space During Automatic Generation of Kalman Filter Code
British Library Conference Proceedings | 2005
|Automatic Certification of Kalman Filters for Reliable Code Generation
British Library Conference Proceedings | 2005
|Wiley | 2000
|