Non Deterministic Planning with Evidence and Paradoxical Reasoning Theories

Paper number

IAC-06-A5.P.02

Author

Mr. Matteo Ceriotti, University of Glasgow, United Kingdom

Coauthor

Dr. Massimiliano Vasile, University of Glasgow, United Kingdom

Coauthor

Mr. Mauro Massari, Politecnico di Milano, Italy

Coauthor

Mr. Giovanni Giardini, Politecnico di Milano, Italy

Year

2006

Abstract

Autonomy is an important feature for space systems, especially for planetary exploration rovers: telecommunication difficulties make appreciable that the system would be able to decide autonomously the best actions to achieve mission goals. Furthermore, for every rover activity, there is intrinsic uncertainty on activity duration, position of the rover, and other environment characteristics that affect each operation, like soil condition, dust on solar panels, temperature, etc.: disregarding them in planning would bring unreliable plans, that are likely to fail.
Most non-deterministic planners (e.g. Buridan, Maxplan, Puccini) have not been developed for the space problem, and so they have strong limitations: actions are instantaneous and have a limited number of possible effects. Moreover, a limited number of possible states are admitted. Remote Agent is the only space planner (on the Deep Space One spacecraft), but it is fully deterministic, allowing replanning in case of plan failure. Nasa Mars rovers (Sojourner and MERs) have no planners, but the latter are able to choose the best path autonomously.
In this paper, a novel, non-deterministic planning approach for autonomous planetary exploration rovers will be presented. Uncertainties in modelling the surrounding environment and in the input from sensors are integrated in the planning process in order to make the rover activity more reliable and to prevent failures. For each plan created by a planner a measure of reliability is computed and used to predict and select the safest one. The evaluation of the plan has been performed with the Dempster-Shafer Theory of Evidence, that allows to deal with both aleatory and epistemic uncertainties. The information regarding the time-length of each action in the plan, the available electric power, the amount of stored energy, the available computer memory, the flow of data sent to Earth, the temperature of the rover, and the Digital Elevation Map of the environment, have been considered incomplete and affected by error. 
Moreover the treatment of incomplete information allows to increase the autonomy of rovers. To this aim, an innovative approach has been designed, consisting in the fusion of uncertain payload and navigation information, gathered by the rover during its mission, from the stereo optical and infrared cameras (the payload). The fusion yields an “interest map”, that quantify the level of interest of each zone around the rover. In this way the planner can choose the most interesting scientific objectives to be analysed, without human intervention, and transform its goals autonomously. The novel Dezert-Smarandache theory of plausible and paradoxical reasoning has been used for information fusion: this theory allows to deal with vague and conflicting data.
Finally the paper shows some sample case of the proposed approach applied to plans, with similar resources and objectives, but different levels of reliability. In this case a Montecarlo simulation has been used to verify the results. Moreover these tests demonstrate how the planner is able to generate plans that maximise at the same time reliability and the level of interest for an artificially created environment.

Abstract document

IAC-06-A5.P.02.pdf

Manuscript document

IAC-06-A5.P.02.pdf (🔒 authorized access only).

To get the manuscript, please contact IAF Secretariat.