Task Switching Model for Acceleration Control of Multi-DOF Manipulator Using Behavior Trees

In recent years, there has been a growing interest in developing robots capable of replacing human labor, driven by factors such as the decline in the working-age population. One crucial aspect of developing such robots is the ability to decompose complex tasks into manageable subtasks and establish models that govern the switching between these subtasks. Finite state machines (FSMs) and behavior trees (BTs) are two commonly used models for task-switching in robotics. FSMs are mathematical models that describe the behavior of a system with a finite number of states. They have been extensively employed in various robotic applications, including gait pattern, trajectory, and motion generation of robots. However, these transition models primarily focus on the relationships at the higher level. Additionally, FSMs are rarely utilized in acceleration control systems, which offer precise position and velocity control, as well as flexible force control and hybrid control capabilities. BTs, on the other hand, are graphical models that represent an agent's behavior as a graph composed of nodes and edges. BTs also enable modelling of switching between multiple tasks and have been explored for automatically generating behavior trees through machine learning techniques. This approach is well-suited for motion control using acceleration control. In this research, by combining BT-based task-switching with acceleration control architecture, we enable autonomous switching between target object approaching and obstacle avoidance. The proposed method is validated through simulations and experiments using a 6-DOF manipulator.