About this Research Topic
The research field of mobile manipulation is at an exciting stage. Researchers have endowed robots with sophisticated motion capabilities. Those capabilities are the prerequisites for robots to perform real-world actions like cooking meals, assembling products, helping dress people, or doing laundry. However, when it comes to endowing robots with the cognitive capabilities to flexibly and competently employ these motion capabilities there are still many open research questions.
A key cognitive capability for robots is ‘manipulation intelligence’ with which we refer to the ability to understand the interplay of motions and effects. To clarify this description, let us consider the instructive example of a robot that shall flip a pancake with a spatula. A robot with manipulation intelligence knows that pancakes can break or fold during flipping. This knowledge allows the creation of an automated feedback loop: Before performing an intended motion, the robot can predict probable effects on the pancake. If those are problematic, e.g. because the pancake appears to be unusually thin, the robot can modify the motion– without the need for human intervention. During and after flipping, the robot can monitor what happens to the pancake and trigger motion adaptation, if unwanted effects like folding occur or the robot failed to flip the pancake.
Obviously, manipulation intelligence combines various cognitive capabilities. For instance, the control systems generating robot motions typically require action- and context-specific mathematical models. Unfortunately, it is still an open question how to autonomously translate given manipulation problems and perceived geometric scenes into meaningful mathematical models for motion generation. Consequently, endowing robots with new manipulation skills or transferring skills to new contexts requires human creative input. Understanding, and eventually automating this cognitive process is just one of the research challenges ahead. Examples of further required cognitive capabilities are prediction and perception of motion effects, understanding of tool affordances, and experience-based motion optimization.
Research on this new generation of robot control systems has already started. As a result, the development of robotic applications will be faster and easier, enabling projects of much grander scale and ambition than the robotics community can tackle today. The resulting robots will possess better abilities to learn new actions and perform known actions in new contexts.
This Research Topic aims to collect the latest research results from scientists that are fascinated and driven by the question of how to build this next generation of robots. Specifically, we invite researchers working on robot control, machine learning, task and motion planning, and knowledge-based robotics who try to combine these technologies. The main objectives of this Research Topic are to disseminate the current state of the art, but also to formulate key research questions, and outline a road map for young researchers to follow.
This Research Topic is based on outputs from the workshop 'Towards Robots that Exhibit Manipulation Intelligence' at IROS 2018. Of course, we welcome submissions not associated with this workshop, assuming they fit the scope of the Research Topic.
We thank Georg Bartels for his substantial contributions in scoping this Research Topic.
Keywords: Mobile Manipulation, Knowledge Representation and Reasoning for Robotics, Learning and Adaptive Systems, Perception for Grasping and Manipulation, Manipulation Planning
Important Note: All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.