As advancements in robot algorithms for manipulation and navigation continue and robot hardware becomes more durable and accessible, there is a growing demand for robots to perform more complex tasks in various settings such as homes and factories. In the past, direct teleoperation, where humans directly control the robots, was the most common and traditional form of control. However, due to the intricate nature of robot motion, human operators often find themselves focusing primarily on solving low-level motion control tasks, which can be mentally taxing.
In this research, we propose a goal-directed approach to programming robots. Our aim is to provide a tool that allows operators to model the world and define goal states for specific tasks. By using three-dimensional (3D) templates on point clouds, operators can set the initial positions of objects and their affordances, as well as their desired goal positions. The robots will then utilize a combination of task and motion planning algorithms to solve the given task.
For more detailed information, please refer to the original paper.