A Historical Perspective
by Oliver Brock (March 2012)
Mobile manipulation is fast-growing area of robotics research. The starting point of this paper is the view that mobile manipulation differs from traditional research areas in robotics, such as perception, motion planning, manipulation, control, cognition, and artificial intelligence. These traditional areas develop mathematical, algorithmic, or practical tools to enable specific skills. These skills are then demonstrated, tested, and evaluated in the context of a specific task or application. In contrast, mobile manipulation seeks to develop integrated robotic systems capable of addressing a wide range of applications in real-world environments. Research in mobile manipulation requires the integration of many if not all traditional research areas of robotics. It emphasizes the development of integrated systems with robust, autonomous, and general capabilities to be performed in the real world, outside the controlled lab environment.
To understand what mobile manipulation is exactly, we should begin by considering the history of this research area. The term mobile manipulation itself might be misleading. It was coined in the nineties, when research labs began to mount robot manipulators on mobile platforms. Back then, the term was intended to capture just that: research conducted on an experimental platform that combines capabilities in mobility and manipulation.
It soon became obvious that combining mobility to manipulation was a game changer. Mobility enables manipulators to leave the lab. They now have to face the complexities of the real world. And in the real world, much of the research that had been successful in the carefully controlled environment of the research lab proved to be brittle or inadequate. This traditional research often relies on substantial prior knowledge about the environment to be successful; it requires a carefully specified experimental environment—a stark contrast to the unpredictability and non-stationarity of everyday environments.
In the context of mobile manipulation research, controlled environments, such as factory floors and specific experimental setups, are often referred to as structured environments. In contrast, unstructured environments are environments that have not been modified specifically to facilitate the execution of a task by a robot. These two types of environments are, of course, opposites on a continuous scale with many possible intermediates. And it must be noted that even unstructured environments contain a significant amount of structure that can be exploited by the robot.
In unstructured environments, it becomes a necessity for the robot to perform a wide variety of tasks, instead of a single specific task. Any specific task, such as retrieving a book from the library, might require the robot to solve additional challenges, such as opening doors, operating elevators, moving a chair out of the way to approach the shelf, or asking for the location of the stacks. Today, no robotic system possesses such capabilities.
Researchers in mobile manipulation have learned that the progress achieved in traditional research areas of robotics cannot simply be combined to produce the level of competency they aim for. And so the combination of manipulation and mobility lead to the founding of mobile manipulation as a research area. The research in this area seeks to develop robotic systems capable of autonomous task execution in unstructured or minimally structured environments. Clearly, the transition from single-purpose lab demonstration—a common practice in robotics—towards such mobile manipulation systems is a gradual one. Mobile manipulation research therefore gradually increases the autonomy, robustness, and task-generality of robotic systems while at the same time gradually reducing the dependence on prior information about the environment and the task.
The objective of mobile manipulation research is to
- maximize task generality, and to
- minimize the dependence on prior information
of autonomous robotic systems.