Autonomous goal management for advanced exploration rovers

Paper number

IAC-08.A3.I.17

Author

Dr. Michèle Lavagna, Politecnico di Milano, Italy

Coauthor

Mr. Damiano Bardella, Italy

Year

2008

Abstract

State-of-the-art rovers, such as NASA MERs, and current on going projects, such as ESA ExoMars,
upload from Ground high level goals to be satisfied within the incoming scheduling timeframe; more specifically, locations of their next science targets are selected on Ground and uploaded via radio contacts. If a space system were able to autonomously select the next goal to refer the next schedule to,
no more waiting for a visibility window to get high-level instructions about its future tasks efficiency of its science would greatly be improved together with its robustness. This paper proposes a possible approach to design a high-level layer devoted to the autonomous on-board generation and activation of goals to be delivered to the system lower control levels. The proposed approach is general but here specifically proposed for exploration rover scenarios. This module, hereafter called Goal Manager (GM), is being developed to be included in the agent architecture developed at DIA of Politecnico di Milano, but it is meant to be widely applicable. Literature highlights some researches are on going on the topic: in [5]
science autonomy is described as the ability of the rover to reason about science goals and the science data it collects to make more effective exploration decisions.In [3] autonomous goal creation is proposed to extend the list of system coordination capabilities. [1] proposes a science event manager to process science and generate an observation request. The solution to the goal creation problem outlined in [4]
makes use of motivations: an agent lead by a set of motives is described, suggesting splitting goal creation in two phases: generation and activation. The here proposed GM is thought to be part of the agent architecture presented in [2]: the autonomous agent architecture focused on merging the deliberative mechanism together with reactive behaviors to cover as far as possible reasoning
mechanisms, the goal selection too. The GM layer is inserted at the top of the agent architecture. This module is based on
motives that cause the agent to act; a motive is a set of options where each option is defined by specific state variables instantiations encapsulating the correspondent motive (behavior class). A set of goals is generated by merging options coming from different motives. The composition procedure relies on a combinatorial optimization problem. Each option contribution to the optimization criterion is not static but it updates according to feedback information from lower layers. This ensures a dynamic goal generationactivation. An ad hoc formalism was created to define motives, options, and
motivations. The GM core is the generation module. The baseline solution is
to limit the number of goals during generation and accept as valid the first goal that is properly planned and scheduled. This is the more suitable solution for general applications. The interpreter module has to change the shape of motivation intensity functions, according to the sensed data, i.e. beliefs. The relationships between beliefs and function parameters have to be a priori set, but a learning mechanism might be considered as future enhancement. Simulation results are here presented on a space exploration rover scenario as the ExoMars should be. In this scenario, relationships describing motivation intensities are position (x,y) and domain dependent physical parameters. Motivation intensities trends have been tuned thanks to a dedicated testing campaign. Simulation results from more complex scenarios are offered, to show the potential of the proposed promising approach to the autonomous goal management problem. The GM layer is currently under testing on the overall agent architecture.

References

[1] S. Chien, B. Cichy, A. Davies, D. Tran, G. Rabideau, R. Castano, R. Sherwood, D. Mandl, S. Frye, S.Shulman, J. Jones, and S. Grosvenor. An autonomous earth-observing sensorweb. IEEE Intelligent Systems, 20(3):16–24, 2005.
[2] M.Lavagna, G. Sangiovanni, A.Brambilla, Autonomous agents in space: hybrid architecture to fulfill deliberative and reactive performance, 9. ESA Workshop on Advanced Space Technologies for Robotics and Automation ASTRA 2006. Noordwijk, The Netherland. November 28-30, 2006
[3] R. A. Morris, J. L. Dungan, and J. L. Bresina. An information infrastructure for coordinating earth science observations. smc-it, pages 397–404, 2006.
[4] T. J. Norman. Motivation-based direction of planning attention in agents with goal  autonomy. PhD thesis, Department of Computer Science, University College London, 1997. url:
citeseer.ist.psu.edu/norman97motivationbased.html.
[5] T. Smith. Rover science autonomy: probabilistic planning for science-aware exploration. Thesis Proposal at Carnegie Mellon University, 2004. url:citeseer.ist.psu.edu/smith04rover.html.

Abstract document

IAC-08.A3.I.17.pdf

Manuscript document

IAC-08.A3.I.17.pdf (🔒 authorized access only).

To get the manuscript, please contact IAF Secretariat.