We have been developing a number of autonomous robots that acquire the strategies to survive in real world outdoor environments. The survival ability refers to the ability of the robots to autonomously maintain their energy amount to a certain level that ensures their ability for executing their strategies. Most creatures can survive by their own decision scheme with several autonomous functions and adaptation mechanisms under an unknown environment. In order to realize these abilities artificially, many approaches have been introduced in various research fields, such as artificial intelligence and biomechatronics, including humanoid robotics. In this research, we are focusing on how the robot can self-implement such functions and what kind of ability is necessary to support the implementation.

The robot uses energy balance as the evaluation function. We consider that the energy balance will be the universal evaluation function for the robot. We introduce the amount of input and output of current fluctuation of the battery, to acquire the amounts of energy consumption and power generation for every action. The proposed adaptation mechanism gives the robot the learning ability to survive in an arbitrary environment through energy-preservation. Natural creatures realize the self-preservation ability by having autonomous and self-reliance strategies. In a similar way, our robot achieves self-preservation by selecting strategies and learning through the interaction between environment and internal state in a real environment.