A robot simulation is provided, which was created with the Box2d physics engine. The human operator can rotate the robot around it's current position, he can open the gripper and it's possible to move the robot forward and backward. The task is to collect the rectangular box and bring it to the base. The good news is, that the teleoperation task can be handled easily with a human in the loop. The simulation is running with smooth 30 fps and the situation is equal to a fork lift truck.
The only problem occurs if the robot should run autonomously. The idea is to program a task planner which works as a text adventure. The world state, possible actions and relevant waypoints are converted into natural language and the planner is solving a quest. For example, the current worldstate contains of a fact, gripper=”close”, but many more information are needed in the form of textual strings.
Unfortunately, the current AI literature doesn't provide much help how to convert a scenario like the puck collecting robot into a text adventure. Sometimes, it was mentioned that expert systems with rules are a good starting point.