Planning for Robotic Systems
Automated Task Planning with Artificial Intelligence
Every scenario in which a team of different agents, robots and humans, are involved, requires a software module that organizes their tasks such that some goals are reached. Task Planning algorithms are developed exactly to achieve this aim. They define the tasks, order them, and assign them to the involved agents, such that at the end of the tasks’ execution, set goals will be achieved. In the task allocation process, optimization aims, such as minimizing the total execution time, are considered. The generated tasks are then executed by a series of basic actions, such as trajectory planning and execution for a serial arm, localization and navigation for a mobile platform, or different processes that must be executed with a gripper.
The real world is dynamic, and elements of it are characterized by a high degree of uncertainty. In such environments, planned actions might not be executed as expected and they can fail. When failures occur, a new plan must be generated and executed. The Automated Task Planning is a special class of task planning algorithms which gives the system the needed autonomy to plan and recover also when tasks fail. At IGMR, intensive research is done to develop a complex Task Planning Framework for teams of cooperating agents around Automated Task Planning methods. This task planning framework is based on the ROSPlan framework and was already validated in a simulated industry scenario and several real-world scenarios.
Funding
Scenarios and Use Cases
Work Range of Mobile Robots
Graph-based planning of construction tasks
Motivation
The automation of tasks in the construction industry is subject to a high economic pressure and at the same time to a high degree of planning inaccuracy, which is why a time-optimized and equally changeable planning method is required
Goal
Development of a planning algorithm that reliably assigns the area-related tasks in the building industry on walls, ceilings and floors for different apartment floor plans for a fleet of robots of any size in order to complete the process as time efficient as possible
Approach
The planning problem is transferred from the given floor plan into a topologically multilevel graph and solved layer by layer by for each individual in the robot fleet by using different optimization approaches. The calculated costs are compared in an auction process and the tasks are assigned to the individuals. The planning loop continues to run continuously even while the work is being carried out, so that sensible changes to the plan can be incorporated at any time.
Partner
This planning module was developed as part of the Bots2ReC research project and was designed in particular according to the requirements of our French partner Bouygues Construction.
Funding
Online Motion Planning and Control
Agile and freely networked assembly systems, which are characterised by the sensoric-supported cooperation of several mobile and stationary robots, are an essential part of the factories of the future.
The dynamic transformation of production lines requires special control system for robotic manipulators.
In current industrial applications, such systems are usually used for recurring tasks and are taught to perform their tasks, thus defining fixed movements. In future assembly scenarios, this is not the case. New requirements ensure that a fast and reliable motion planning and control scheme must be developed. This scheme must respond appropriately to changes changes in the environment and setpoints and produces feasible motions. These modern assembly systems can then be enabled to become a part of the Internet of Production by being placed in the structure of a meta-model, the so-called digital shadows.
As part of the Cluster of Excellence” Internet of Production”, we model such assembly scenarios via coordination and consensus in a sensor-robot network.
Funding
Cluster of Excellence Internet of Production
Selected Publications