Guest Editorial Special Section on Soft Computing for Space Autonomy

Robotic and human exploration of our solar system requires space missions to be designed with high levels of autonomy for dealing with hostile environments and restricted communication links. Researchers are discovering new solutions to associated problems using fuzzy logic, artificial neural networks, evolutionary computation and probabilistic reasoning, some of the principal elements of soft computing. The uncertainty of space environments and the need for safe mission operations in response to anomalies makes such soft computing techniques good candidates to improve the development of " intelligent autonomous systems " needed for future space missions. This special section provides a representative cross-section of innovative and practical research and applications of Soft Computing for Space Autonomy. The articles offer recent developments in soft computing for robotic applications, computer vision, control, monitoring, and scheduling. The special section begins with an article by Remy and Howard presenting a methodology that enables a robotic system to interactively learn how to perform tasks from examples provided via teleoperation by an astronaut, mission designer, or engineer. The methodology employs self-organizing maps in the form of neural networks to extend a robot's ability to operate with a competence on par with its human teacher, thus enhancing human-robot interaction potential for space missions. Representative application examples in relevant task scenarios are presented and neuro-controller, classical controller, and human teleoperator task performance are measured and compared. Massari, Sangiovanni, and Bernelli-Zazzera provide a detailed investigation of properties of a dynamic neural network controller for a six-legged rover prototype designed for rough planetary terrain mobility. The controller is implemented as a continuous-time recurrent neural network designed with the aid of evolutionary algorithms. Effectiveness and robustness of the controller are demonstrated by convincing experimental results in simulation and on a physical rover testbed considering sensor errors and failure as well as perturbations in neural network parameters. Their approach enables soft computing-based dynamical control of multi-degree of freedom systems and achieves excellent results while circumventing more complicated model-based control design methodologies. The article by Santos et al proposes automatic generation of fuzzy sets for fault monitoring and terrain recognition tasks associated with a drilling system for the ESA ExoMars rover. A novel sensor-driven fuzzy set learning method is proposed as a time-efficient alternative to