Artificial Intelligence for the Development of a Robotics Ecosystem for Design and Manufacturing
Mobile, Intelligent and Autonomous Systems Are Becoming Reality
- Fig. 1: Humans and robots work hand in hand supported by digital assistants. (Source: DFKI GmbH, © Uwe Völkner / Photo Agency)
- Fig. 2: Overview of the links in the D-Rock database - hardware, software and simple behavior can be stored here (Source: DFKI GmbH, see also https://www.youtube.com/watch?v=DwqY3hsGlU8&feature)
- Figure 3: Q-Rock - From Hardware to Behavior (Source: DFKI GmbH)
More and more powerful sensors and actuators are turning the vision of highly mobile, intelligent and autonomous systems into a reality. As a result, the complexity of robot systems is steadily rising, thus developers are facing ever greater challenges to cope with this complexity. Acoordingly, the development costs are increasing, making the design of customized robots for specific applications especially difficult for small and medium-sized enterprises.
Launched in August 2018, the Q-Rock project, funded by the German Federal Ministry of Education and Research (BMBF / FKZ 01IW18003), is taking a revolutionary approach to circumvent these problems: With the help of Artificial Intelligence (AI) methods, it will be possible for future non-expert users to develop robot systems for their applications in a cost-effective manner.
Artificial intelligence opens up new application fields in robotics
The current rapid developments in the hardware and software of robotics as well as AI are leading to ever new fields of application for robots. This will affect almost all areas of human life. In particular, robots should work hand-in-hand with humans in the future (fig. 1) and become an inherent part, both in work life and in private households. Initial projects are already dealing with the architecture of so-called hybrid teams of autonomous robots and humans [1, 2]. At the same time, high technological requirements have to be fulfilled if a robotic system has to operate autonomously supporting people in their living environment: both hardware and software become highly complex and no longer develop independently of each other. In order to interact with their environment, humans and possibly also other robots, the systems must be equipped with appropriate interfaces and need to have a high level of flexibility. It is already foreseeable today that the complexity of these developments can and will only be overviewed by a few experts. Frequently, entire teams of experts work on system development and programming, while the diversity of the components is simultaneously increasing. In science, therefore, new approaches are sought to both reduce the complexity of the design process  and to categorize the behavior of robots with the help of algorithms .
The technology should not only benefit a few, instead a broad social acceptance and competence in dealing with robotics and AI is needed.
However, it is clear that not everyone can become an expert in the development of robots. Instead, it is necessary to reduce the required skills for the development of robots as much as possible, so that one can construct a robot with the background knowledge of the respective field of application. To achieve this is the core idea of the X-Rock projects, which are carried out at the Robotics Innovation Center of the German Research Center for Artificial Intelligence. The idea is to coordinate, abstract and automate the co-development of hardware and software as a first step. For this purpose, various AI techniques are used, which help the developer and support robot development with knowledge and intelligent processes.
Linking database, hardware and software
The central element binding together all the developments is a database that links hardware components with software components and other specifications. Enrichment of data is ensured by compatible software for development and operation. As a result, the expert knowledge for already existing and devised systems should be included in the database. For example, if a robotic arm has been developed, then it is part of the database with its hardware, firmware and software. The next user, who needs an arm and wants to develop it further, can simply load and reuse the already existing data. The basis for such a database and the concatenation of appropriate development software was laid in the project D-Rock (funded by the BMBF, FKZ 01IW15001), which was completed in 2018. The link is shown in figure 2.
The recently launched project Q-Rock is about the next level, namely the behavior of a complex system. The project assumes that users of a robot will only want to specify a list of requirements in the future. It allows users to describe the capabilities of a system without having to deal with the technical details of design and manufacturing.
The goal of Q-Rock is to determine the connection between hardware and behavior (see Figure 3) supported by AI, based on the principle that the skills of a system are based on its hardware. For instance, the operational capability of an arm is amongst others defined by the working space, which again is based on the arrangement of the joints and actuators. It is of crucial importance here to determine how a robot can develop the knowledge about itself and its abilities on its own, without having a definition by a developer.
The approach in the project is using a three-step development cycle, as shown in Figure 3. The cycle begins with the exploration of capabilities by the robot based on the information available about the hardware (and possibly software). These can be calculated to a limited extent by analytical methods. Moreover, deep machine learning AI methods using a simulation should explore the capabilities from the hardware. This exploration is used to determine all accessible states of sensors and actuators in the system and their relationships to the states and actions of the robot. Such a "bottom-up" exploration should also be able to discover completely new hardware capabilities that users did not initially think about.
In the second part of the cycle, a classification of the explored capabilities towards building behaviors takes place. The capabilities are grouped and semantically annotated using expert knowledge or automated approaches. The results are then stored in the database. The semantic annotation is created by direct interaction with the user and thus contains concepts from his/her perspective. As a result, the database contains a description of the hardware, showing its skills from a user perspective.
Holistic solution for all users
In order to finally be able to specify a robot by specifying its intended behavior, a reasoning structure is set up in the third part of the cycle. For this purpose, the annotated skills are mapped onto a behavior with the help of a hierarchical hybrid planning and hence a direct connection for the application is established to both, hardware and behavior. Based on this information, the Q-Rock system can suggest a robot to the user. If something has to be added and if it is not already part of the database, the system can then generate and complete missing data by re-running the Q-Rock development cycle.
The resulting system of the Q-Rock project is therefore intended to be a further step towards a holistic solution enabling users to develop robot systems customized to their own needs using a modular construction principle. The aim is to make the complexity of this endeavor manageable to the extent that the public and especially small and medium-sized companies have broad access to robotic solutions in order to be able to keep up with the development speed of market-dominant players.
In addition, Q-Rock will alter the procedure of robot development in several ways: by enabling the automated design of robotic hardware based on the desired behavior, completely new design and planning processes for robot applications will become feasible in the future. In addition, it will be possible with the help of Q-Rock to conclude in a model-based fashion which tasks a robot can perform with its given hardware through compositions of behavior. Thereby the results should also be applicable for the qualification of hardware.
The results from the X-Rock project line will be gradually made available in an open source platform. The provider of the platform is currently planned to be the DFKI spin-off Raise Robotics (www.raise-robotics.com)
Sirko Straube1, Hendrik Wöhrle1, Frank Kirchner1
1Robotics Innovation Center, Deutsches Forschungszentrum für Künstliche Intelligenz GmbH, Bremen, Germany
 T. Schwartz, H.U. Krieger, I. Zinnikus, C. Bürckert, J. Folz, B. Kiefer, P. Hevesi, C. Lüth, G. Pirkl, T. Spieldenner, N. Schmitz, M. Wirkus, and S. Straube: Hybrid Teams: Flexible Collaboration Between Humans, Robots and Virtual Agents. In Proceedings of the 14th German Conference on Multiagent System Technologies, (Mates 2016), Klagenfurt, 2016
 Straube, S., Schwartz, T.: Hybrid teams in the digital network of the future -application, architecture and communication. Industrie 4.0 Management 2, pp. 41-45, 2016.
 Ha, S., Coros, S., Alspach, A., Bern, J.M., Kim, J. and Yamane, K.: Computational Design of Robotic Devices From High-Level Motion Specifications. In IEEE Transactions on Robotics, vol. 34, no. 5, pp. 1240-1251, 2018.
 Zamir, A.R., Sax, A., Shen, W., Guibas, L.J., Malik, J., and Savarese, S.: Taskonomy: Disentangling Task Transfer Learning. In Proceedings of The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2018