Teams of the Robotics Innovation Center
The long term goal of the topic “Long Term Autonomy”, which is handled by the Robotics Innovation Center (RIC) of the German Research Center for Artificial Intelligence, is to answer the elemental question, how intelligent behavior of technical systems in complex and dynamical environments can be achieved sustainably for long periods of time, i.e. for months and years.
Starting point for this approach is the assumption that only systems which feature a minimum of structural complexity even possess a disposition for a sustainable interaction with natural environments and thus represent appropriate platforms for control and feedback control systems.
At DFKI RIC, six teams of researchers are working together on the realization of this vision: The team System Design is focused on the fields of construction and integration of robot systems, based on the most recent production technologies and materials. The team Hardware Architectures is using, however, the latest chip technologies to develop and integrate parallel, decentralized, and fault-tolerant networks of embedded systems. Adaptive control and real-time simulation strategies and methodologies for complex robots are the topics the team Behavior Control & Simulation deals with.
On the other hand, the team Sustained Interaction and Learning is developing methods that allow a robot to independently identify new possible solutions for given tasks and to autonomously acquire the structure of complex tasks. In addition, this team focuses on the human, being part of an increasingly connected, exploited world, which is permeated with robotic systems. In this context, the team develops non-invasive methods for an optimal, user-adaptive integration of the human being in the interaction loops, based on methods of artificial intelligence.
The team Robot Control addresses feedback control related aspects of the robot and a compliant control for complex, closed, and parallel kinematic chains.
Finally, the team Knowledge-based long-term Autonomy develops tools and methods that enable robots to navigate autonomously for long periods of time in environments unknown to them.
The objective of team “System Design” is the development and assembly of innovative mechatronic systems as a basis for intelligent and autonomous robots. Mechatronic systems are characterized by a fusion of mechanical and electronic components as a precondition for capable robotic systems. Balancing the contrary requirements of mechanical strength versus lightweight design while simultaneously using the mechanical structure as an element of a measurement chain to determine the internal system state are challenges to the design of robotic systems.
A fundamental paradigm in the design process of mobile robots is lightweight construction. To implement the principles of lightweight design in systems with high dynamic abilities, the DFKI takes advantage of modern materials and production technologies, as well as functional integration. Smart materials are used as sensors, actuators and parts of the framework to increase the robots performance.
One focus of the team “System Design” is modularization and the development of electromechanical interfaces. Modular systems allow a fast adaption to changing boundary conditions, by reusing known solutions for an efficient development process.
Head of team: Dr. Marc Simnofske
The independent and long term interaction of technical and especially robotic systems with their environment is based on sensor data. To cope with the increasing amount of collected sensor information in the next-generation robotic systems, the team develops appropriate processing architectures. Therefore, next generation chip technologies and approaches from artificial intelligence are used already at chip level. For the use of robots for long-term missions, the hardware must be robust and fault tolerant. To achieve this, the team investigates suitable reconfiguration techniques as well as self-healing approaches. Another research topic of the team is the perception of energy consumption on component-level, which paves the way of energy-aware mission planning. The energy efficient design of systems is continuously accompanying the development of our robots.
Additionally, the team is working on the realization of highly integrated sensor solutions. Driven by the desire to capture the environment of a robotic system as detailed as possible, the density of sensing devices of different modalities is increasing. This requires novel approaches to sensor integration. Among other topics, the fusion of different measuring principles, the processing of the collected information on site as well as the integration of signal conversion and processing is part of the work of the team “Hardware Architectures”.
Head of team: Dr.-Ing. Peter Kampmann
The team „Robot Control“ deals with the development of strategies to control dynamical systems ranging from a single robot actuator to highly complex whole-body robot morphologies such as those of humanoids and mobile manipulators. Focus is given to a change of paradigm from a classical pure kinematic to a more complex dynamic control which takes into account robot and environment dynamics to enhance physical performance and allow compliant contact interactions. In cooperation with the Teams “Sustained Interaction & Learning” and “Behavior Control & Simulation” the research efforts of the team “Robot Control” also focus on the increased use of multimodal sensors for reactive control as well as for recognition and prediction of contact events. Another important challenge that will be tackled in cooperation with the team “Autonomy” is to integrate the control approaches into the overall frameworks and architectures.
Future robotic systems will possess highly complex kinematics which will require holistic strategies such as whole-body control for dealing with the large number of joints and the need to control multiple and simultaneous robot tasks. For instance, a humanoid robot on two legs on top of a ladder drilling a hole in the wall has to keep its body posture and balance under control while ensuring a successful and accurate drilling. This goal, in turn, will require efficient dynamics identification strategies as well as efficient kinematics and dynamics algorithms. Optimal control and the use of predictive strategies are also required both at actuator and body level in order to be robust, fault-tolerant and adaptive on different environments and tasks. Similarly, once the robotic systems interact with humans, ensuring compliance and using contact and collision detection algorithms are of utmost importance.
In summary, the team “Robot Control” is mainly concerned with the control and the dynamics of the movement apparatus of highly complex robots such as humanoids walking on two legs. The goal is to develop generic robot-independent control strategies which take into account environmental and contextual information to adapt to diverse environments. In collaboration with the other teams, the systems are designed to operate in highly unstructured and dynamically changing scenarios.
Head of team: Dr.-Ing. José de Gea Fernández
Behavior Control & Simulation
Due to the progressing technological developments, we are able to continuously increase the sensory and motor integration density and equipment of robots to give the systems more extensive abilities to perceive their environment and to act in it in different ways. To exploit the gained capabilities to the full extent behavior control of such complex systems is becoming more challenging. For the generation, testing, evaluation and optimization of various behaviors, tools for the simulation of robots and their interaction with the environment are indispensable.
The team deals with behavior control of robots with extensive sensorimotor disposition. In this area, efficiency, flexibility, and adaptability of the robot should be increased in order to improve their skills of manipulation and mobility. Other requirements such as high stability, robustness and reliability play an important role. The research and development efforts also focus on the increased use of integrated sensors, which will be developed in cooperation with the team "Hardware Architectures", in motion control of new kinds of actuators and innovative kinematics, which are conceived in collaboration with the team "System Design".
Systems, which will be used in the future, must be able to adapt to different environments, requirements, and tasks – but must also be sufficiently robust. Once the systems interact with humans, or are used in hazardous environments, highest safety requirements result in special challenges to research and technology development. On the implementation of such control concepts we are working together with the team "Robot Control".
Depending on the application scenario, the systems should be able to act autonomously and reliably over long periods of time. A particular challenge in terms of long-term autonomy of robots is to adjust the control parameters to changing system parameters. For example, a joint having an increased friction due to wear should be stressed less. The robot has to be able to recognize such a malfunction autonomously and to identify and use alternative movement strategies. Since nature itself is the best model for such systems the intended approach is to transfer skills, functional principles, morphologies and control approaches from biological to technical systems.
Tools for the simulation are used to support the development and operation of robotic systems or certain procedures such as the use of machine learning methods as it takes place in the team "Sustainable Learning and Interaction". Thus FEM-simulations for optimization of flow bodies of underwater robots as well as Rigid-Body-Simulations for optimization of the walking robots’ morphology and behaviors are applied. One focus is concentrated on real-time capable simulations which enable development and testing of the system’s software. In addition, these simulations can be directly integrated into the control software of robots and used at runtime to e.g. evaluate the success and the potential consequences of the intended action before execution. Thereby, we can increase the capabilities of autonomous systems can together with the team "Autonomy".
Further issues are: “Hardware in the Loop” (HIL) which means integration of real hardware in a simulation system, e.g. for testing sensors, “Virtual Reality” for virtual representation and for testing of robots in their designated environment, immersive display for the operator via head-mounted displays or in a CAVE-like environment for better, simultaneous control and monitoring of (several) systems.
In addition to application of these issues, the scientists are working on their further development. Thus, methods are developed in order to integrate new, not previously existing features and functions in the real-time simulation and to minimize the gap between simulation and reality.
Head of team: Dr.-Ing. Sebastian Bartsch
Sustained Interaction & Learning
The vision of the team “Sustained Interaction and Learning” are interactive, autonomous robotic systems which can act and learn in a complex environment over a longer period of time. Without requiring any assistance or external tasks from human supervisors, the systems are able to intuitively interact with humans, to support them and learn from them in various situations. Thereby, robots will not only improve their own behavior, but also will be able to react to different characters of human beings. This enables a sustained collaboration based on the optimal usage of the cognitive skills of the human being while individually being assisted by autonomous robotic systems.
Researchers of the team “Sustained Interaction and Learning” develop new methods that allow autonomous robots to learn self-contained and sustained from their experience, to consolidate the acquired knowledge, and to store this knowledge persistently using appropriate representations. By learning autonomously to model their current surrounding, to adapt their behavior accordingly, and to predict the effects of their action, robots can act in dynamic environments und are able to continuously adapt to them. In addition, the systems’ ability to learn complex behaviors based on basic behavior primitives serves to better understand human behavior and to make an interaction significantly more intuitive. Furthermore, the robot’s behavior can be adapted in such a way that it becomes more predictable for the human, thereby increasing the acceptance of a robotic system as partner in interactions. The predictability of the robot for the human is in turn linked to the predictability of the human for the system, having an important influence on the safety of the human-robot interaction.
The cooperation between the teams “Sustained Interaction and Learning” and “Robot Control” focuses particularly on relevant aspects in the control of robotic systems. Mobile hardware solutions or such which can be integrated into the systems are developed in close cooperation with the team “Hardware Architectures“. Moreover, the team “Sustained Interaction and Learning” has to closely collaborate with the teams, “Knowledge-based long-term Autonomy”, “Behavior Control & Simulation”, and “System Design”.
Head of team: Dr. rer. nat. Elsa Andrea Kirchner
Knowledge-based long-term Autonomy
The central goal of Team Knowledge-based long-term Autonomy is the research of processes and software infrastructures in order to be able to implement robust, long-term autonomous systems in harsh environmental conditions. This requires resilient systems that can react to changes in their own hardware and software as well as to changes in the environmental conditions, and that can adapt their actions to these changes. For this purpose, the systems must capture the semantics of the change and be able to derive the impact on themselves and their tasks. Accordingly, robust and fail-safe robot control architectures, semantic environment perception, the representation of these data, and semantically supported robot capabilities are central research topics of team Knowledge-based long-term Autonomy.
The aforementioned research topics are jointly advanced, since there are strong interdependencies between them, and complex robot behaviors (e.g. active perception or hybrid hierarchical action planning and its execution) can only be realized by combining these topics.
Moreover, team Knowledge-based long-term Autonomy - together with the other teams – works to ensure that contextual knowledge finds its way in all areas of robot control. This makes it possible to consider the relevance, properties, and effects of objects or actions in robot functions such as object recognition, navigation, manipulation, and interaction. Thus robot functions can be developed which – in the respective context - work much more robustly and in a more appropriate manner.
These research topics are accompanied by the development and research of a robust robot control framework which is able to achieve the stated objectives, since currently no framework provides the necessary prerequisites for topics such as model-based robot development, hardware and software configuration, long-term autonomy, fail-safe control structures, and consistent knowledge-based robot control.
With this research focus, team Knowledge-based long-term Autonomy closely collaborates with all other teams of the Robotics Innovation Center:
The hardware developed by team System Design significantly influences the abilities and functions (perception, navigation and planning) implemented by team Knowledge-based long-term Autonomy on this hardware. On the other hand, the tools developed by team Knowledge-based long-term Autonomy for model-based robot development can support and influence the system design processes in the long term.
Based on the work of the team Hardware Architectures, team Knowledge-based long-term Autonomy is able to distribute the control algorithms on the robot. The topics of hardware and software configuration require a close collaboration between the teams. In addition, the new sensor modalities developed here open up further possibilities for team Knowledge-based long-term Autonomy for semantic environment perception.
The team Robot Control enables team Knowledge-based long-term Autonomy to integrate mobile manipulation on an abstract level into the robot’s action planning and execution. In turn, team Knowledge-based long-term Autonomy provides the Robot Control team with the contextual knowledge, enabling them to consider positions, velocities and semantic properties of the relevant objects and persons in manipulation planning.
There is a similar interface between team Knowledge-based long-term Autonomy and team Behavior Control & Simulation, providing team Knowledge-based long-term Autonomy access to the robot’s mobility – independent of its morphology. Here, too, an interface exists to integrate semantic environment information into behavior control.
The team Sustained Interaction & Learning also benefits from the semantic environment representation and the contextual knowledge provided by team Knowledge-based long-term Autonomy, since the interaction with humans can take place more situation-dependent. Using the same vocabulary for humans and robots, interaction can easily be transferred into the robot’s behavior. In addition, the research focus Learning requires close collaboration with team Knowledge-based long-term Autonomy in the fields of environmental perception and semantic environment representation.
Head of team: Dr. rer. nat. Stefan Stiene