High level task planning with inference for the TIAGo robot

Abstract

The need to combine task planning and motion planning in robotics is well understood. The task planner generates a plan to solve the problem while the motion planner executes the actions of the problem. The previous framework is applied in many state machines that solve complex problems. But in this project we want to present an interface that communicates the task planner layer and the motion planner layer, and updates the geometric information of the environment to inform the task planner. This framework allows to solve complex tasks with basic information of the goal, and replan whenever the motion could not be executed. All the information of the problems is modelled as logical predicates. The objective of this project is to generate a generic model of the environment, with a set of feasible motions of the robot, and use this interface to solve many different planning problems involving those actions, by just giving simple goals. The result is to make the robot more autonomous and allow that any user could use it by giving simple orders. Moreover this project presents the different frameworks and algorithms used to simulate those actions in the robot such as: Sequential Quadratic Programming optimization, Rapidly Random Exploring Tree (RRT) or SBPL global planning. It also shows an introduction to PDDL language used to model the problem and the actions, and the Fast-Froward (FF) solver that is the responsible to translate the problem as a graph and solve it. Finally we test it on different experiments in simulation, by using the TIAGo platform of PAL robotics. The results are promising and allow to dream in service robots solving complex tasks simply computing and modelling basic actions.