Abstract: Autonomous robot is a kind of cyber-physical system which operates in an open environment and can make and execute its behavior autonomously.It generates and executes effective plans according to the task.The dynamic change of environment state often results in that the planned behavior strategy is no longer applicable to the current environment and the results of behavior execution are not in line with expectations,thus affecting the task achievement of autonomous robots.The above problems pose challenges to both the behavioral decision-making and the construction of autonomous robot software.On the one hand,autonomous robots need to strengthen the observation of environment states and their changes during the execution of its behavior strategy,and adjust the behavior strategy in a timely and flexible manner based on the observation results,which improves the complexity of the robot’s observation scheme and behavioral decision-making algorithm.On the other hand,the complex interactions of observation,decision-making and behavior execution enhances the complexity of software component abstraction and data interaction,and how to abstract the functions of software components such as sensing,decision-making and actuating of robots and provide suitable software architecture has become an important challenge for the software construction of autonomous robots.In view of the above challenges,the idea of adjoint behavior of autonomous robots is proposed,the adjoint interaction between observation and actuating behavior is clearly defined,and two adjoint observation schemes,conditional adjoint observation scheme and objective adjoint observation scheme,are proposed according to different stages of behavior execution,so as to improve the sensing of environmental changes and decision-making adjustment ability of autonomous robots.Secondly,AutoRobot,an autonomous robot software development framework based on multi-agent system,is developed,which abstracts the robot’s sensors,actuators and planners into a set of autonomous software agents,and the agents implement the above adjoint observation schemes through autonomous decision-making and collaboration.The AutoRobot framework designs and packages a set of reusable software components for different agent types,which can effectively support the reuse and efficient development of autonomous robot software.Finally,an experimental analysis is carried out in the simulation environment,and the efficiency and effectiveness of task planning and execution based on the adjoint observation schemes are verified by comparing it with ROSPlan and DESPOT,two autonomous robot task planning and execution methods.

Key words: Autonomous robots, Open environments, Adjoint observation schemes, Task planning, Task execution