Wojciech Dudek of Warsaw University of Technology

Task management is a core ability of living and artificial autonomous entities. In robotics, it is especially crucial for versatile service robots that are tailored to help humans in various duties. In some applications, robots are shared by multiple users that don’t agree their requests. Versatile service robots often work in dynamic environments, and their users’ needs and preferences change over time. Thus, the robots need an advanced task management ability that includes i.a. dynamic priority assignment, repetitive check of tasks’ feasibility and task plans update. Additionally, robots need to foresee the consequences of their task interruption (e.g. leaving a cooker on). Therefore, the problem of prudent task management respecting the danger of interrupting a robot’s tasks arises. Prudent task management considered in this talk is constituted of i.a. safe suspension and resumption of independent tasks, schedule parameters reappraisal invoked by changes in the environment and termination of a queued tasks that are no longer feasible.

The talk will discuss a model of robot control systems with prudent task management. The model is specified with an agent-based approach and consists of multiple agents of various classes. Each agent class feature a~hierarchical finite-state machine (HFSM) specifying the agent's behaviour. Dynamic Class Agents execute tasks in the model. States of these agents' HFSMs have basic behaviours and the behaviours' terminal conditions assigned. Besides the states responsible for the task execution, some states suspend the current task safely before a task switch. Thanks to that, the model enables mitigation of the problems in predicting the consequences of the robot’s activity interruption.

The model was verified by implementing two example control systems of different robots (TIAGo mobile robot and Velma mobile manipulator). The robots work in different environments, and various algorithms manage their tasks.