DNNF-based belief state estimation
AAAI'06 Proceedings of the 21st national conference on Artificial intelligence - Volume 1
| Hi-index | 0.00 |
There is growing demand for high-reliability embedded systems that operate robustly and autonomously in the presence of tight real-time constraints. For robotic spacecraft, robust plan execution is essential during time-critical mission sequences, due to the very short time available for recovery from anomalies. Traditional approaches to encoding these sequences can lead to brittle behavior under off-nominal execution conditions, due to the high level of complexity in the control specification required to manage the complex spacecraft system interactions. This work describes timed model-based programming, a novel approach for encoding and robustly executing mission-critical spacecraft sequences. The timed model-based programming approach addresses the issues of sequence complexity and unanticipated low-level system interactions by allowing control programs to directly read or write “hidden” states of the plant, that is, states that are not directly observable or controllable. It is then the responsibility of the program's execution kernel to map between hidden states and the plant sensors and control variables. This mapping is performed automatically by a deductive controller using a common-sense plant model, freeing the programmer from the error-prone process of reasoning through a complex set of interactions under a range of possible failure situations. Time is central to the execution of mission-critical sequences; a robust executive must consider time in its control and behavior models, in addition to reactively managing complexity. In timed model-based programming, control programs express goals and constraints in terms of both system state and time. Plant models capture the underlying behavior of the system components, including nominal and off-nominal modes, probabilistic transitions, and timed effects such as state transition latency. The contributions of this work are threefold. First, a semantic specification of the timed model-based programming approach is provided. The execution semantics of a timed model-based program are defined in terms of legal state evolutions of a physical plant, represented as a factored Partially Observable Semi-Markov Decision Process. The second contribution is the definition of graphical and textual languages for encoding timed control programs and plant models. The adoption of a visual programming paradigm allows timed model-based programs to be specified and readily inspected by the systems engineers in charge of designing the mission-critical sequences. The third contribution is the development of a Timed Model-based Executive, which takes as input a timed control program and executes it, using timed plant models to track states, diagnose faults and generate control actions. The Timed Model-based Executive has been implemented and demonstrated on a representative spacecraft scenario for Mars entry, descent and landing. (Copies available exclusively from MIT Libraries, Rm. 14-0551, Cambridge, MA 02139-4307. Ph. 617-253-5668; Fax 617-253-1690.)