Planning in the RoboCup Logistics League

LLSF Simulator Screenshot RoboCup is particularly well-known for its soccer leagues, but there are an increasing number of application leagues. The newest one is the RoboCup Logistics League (RCLL) where groups of robots take on the task of in-factory production logistics. Such systems require a task-level executive, a software component that decides on the actions to take to achieve a certain goal. Typical approaches can be roughly divided in three categories: state machine based controllers like SMACH, XABSL, or the Lua-based Behavior Engine, reasoning systems from Procedural Reasoning Systems or rule-based agents to more formal approaches like Golog, and finally planning systems with varying complexity and modeling requirements. There are also hybrid systems integrating aspects of more categories like integrating PDDL-based planning into Golog.

Planning systems are still the exception rather than the norm in robotics applications, with some notable exceptions. Often domains are either too dynamic requiring prohibitively frequent decision points (e.g., robot soccer), or are highly complex imposing tedious modeling requirements to cover (a suitable subset of) the domain (e.g., domestic service robots).

In this project, we investigate applying planning techniques to the RoboCup Logistics League (RCLL). We find it is a suitable testbed for a wide variety of planning methodologies. It is a medium complex domain inspired by problems from in-factory logistic applications, where a group of robots has to maintain and optimize a material flow among processing machines and eventually deliver to an exit gate. You can get an impression of the Game Play 2014 in the video below.

Related Publications

Simulation

The eventual results are determined on groups of up to three robots based on the Festo Robotino 3. But for development, or to allow application of approaches by teams that do not have actual robots, we have developed our Gazebo-based RCLL simulation. It allows for a quick start into the league and to run the domain as a benchmark, for example before considering to go for the actual robots.

Software Stack Release

We have released the software stack for the RCLL 2014 in full, including our current task-level executive based on the CLIPS rule-based reasoning system as well as all other components required to play the game. We expect to provide an updated release for the 2015 gameplay after RoboCup 2015.