Research

The aim of the Hybris research unit is to integrate both qualitative and quantitative forms of reasoning, resulting in hybrid reasoning formalisms.

In the Hybris C1 project we address hybrid reasoning challenges for a robot planning and acting while dealing with incomplete and uncertain knowledge. We are working on the perception, manipulation, planning, and reasoning aspects of an active perception framework and their integration on real robotic platforms. This enables the robot to reason about the state of its knowledge and to take action if it is still insufficient or too uncertain.

We aim for new hybrid reasoning challenges in the context of service robots operating alongside humans in domestic environments. Specifically, we address robots that attend to tasks according to the preferences of their users during execution. This involves multiple problems such as learning from humans and their environments, dealing with interruptions, and handling complex hybrid planning challenges efficiently.

Spatial Cognition deals with the acquisition, organization, utilization, and revision of knowledge about spatial environments, be it real or abstract, human or machine. Research issues range from the investigation of human spatial cognition to mobile robot navigation.

The Transregional Collaborative Research Center Spatial Cognition investigates the cognitive foundations for human-centered spatial assistance systems. It is structured into the three research areas Reasoning, Action, and Interaction. Reasoning comprises projects that are concerned with internal and external representations of space and with inference processes using these representations. Action comprises projects that deal with the acquisition of information in spatial environments and with actions and behavior in these environments. Interaction comprises projects that address communication about space by means of language and maps.

In a globalized and interconnected business world, production interruptions including supply chain disruption have been the leading business risk for many years.

The ability of a company to permanently adapt to internal and external changes and disruptions is the "quest for resilience". Reinforced by a significant increase in complexity in production due to Industry 4.0, resilience management is thus becoming an indispensable success factor for production companies.

The SPAICER project is developing a data-driven ecosystem based on lifelong, collaborative and low-threshold smart resilience services (SRS) by leveraging leading AI technologies and Industry 4.0 standards with the goal of anticipating disruptions (anticipation) and optimally adapting production plans to active disruptions at any time (reaction).

The KARIS project focuses on a limitation of current intralogistic systems: today, many functionalities are implemented by rigid, permanently deployed "hardware", precluding a flexible, dynamic adaptation to the actual flow of materials. Infrastructure that is designed and produced today requires time-consuming, costly alterations tomorrow, in order to comply with constantly changing requirements.

The development of novel, intelligent and autonomous functional modules for object transportation shall lead to a fundamental change in the design of future intralogistic transportation systems. Rather than using rigidly deployed elements such as roll conveyors, cheap intelligent transportation agents come to the fore whenever an object must change its location.

Complex systems such as airplanes, trains, and cars, contain numerous safety-critical electronic components. Failure of such devices puts lives at risk. For this reason, verifying the correct behavior of electronic components is critically important. Current automatical verification techniques are limited to certain aspects of the overall behavior, such as stability and continuity.

The Transregional Collaborative Research Center AVACS ("Automatic Verification and Analysis of Complex Systems") aims at taking the state of the art in automatic verification to a new level where it is possible to comprehensively analyze the behavior of complex systems and verify their safety properties with automatic techniques.

Within AVACS, our research group focuses on the question how Artificial Intelligence techniques such as heuristic search and game tree analysis can be profitably applied to the automatic verification of hard- and software.

Fast-Forward, abbreviated FF, is a domain-independent planning system developed within our research group by Jörg Hoffmann. FF can deal with arbitary planning problems specified in the standardized input language PDDL (Planning Domain Definition Language). FF can deal with classical (STRIPS) problems as well as full scale ADL problems, including quantification and conditional effects. The Metric-FF planner, which is implemented as an extension to FF, can handle numerical state variables in addition to FF's capabilities. Both systems are implemented in C.

FF won the fully automated track of the Second International Planning Competition (IPC 2) at the Fifth International Conference on Artificial Intelligence Planning and Scheduling (AIPS 2000). It also won the Schindler Award for the best performing planning system in the Miconic-10 Elevator domain, ADL track.

Fast Downward is a domain-independent planning system developed within our research group. Fast Downward can deal with arbitrary planning problems specified in the standardized input language PDDL (Planning Domain Definition Language), supporting all propositional language constructs (STRIPS, ADL, axioms). The system is implemented in Python and C++.

Fast Downward won the "classical" track (propositional problems, no optimization) of the Fourth International Planning Competition (IPC 2004) at the 14th International Conference on Automated Planning and Scheduling (ICAPS 2004). The planner LAMA, implemented on top of Fast Downward by Silvia Richter and Matthias Westphal, won the classical track of the Sixth International Planning Competition (IPC 2008) at the International Conference on Automated Planning and Scheduling (ICAPS 2008).

IPP is a domain-independent planning system developed within our research group by Jana Koehler and others. IPP began as an extension of Blum and Furst's Graphplan algorithm to the ADL language, including conditional effects and expressive preconditions. Later versions included several influential new ideas such as a technique for the removal of irrelevant facts and operators (RIFO) from the planning task and the goal agenda manager (GAM) for decomposing complex planning tasks into much smaller subproblems.

Between July 1997 and December 1998, IPP source code has been downloaded more than 200 times, and the system has been used in several other projects (see e.g. the AIPS 2000 proceedings for references). IPP was also used in two commercial environments: at Schindler for rapid prototyping of elevator domain models (with our permission), and at Celcorp as a source for own planning software development (without our permission).

IPP won the ADL track of the first International Planning Competition (IPC 1) at the Fourth International Conference on Artificial Intelligence Planning Systems (AIPS'98).

The EU-funded project CoSy is a multi-disciplinary investigation of requirements and design options for human-like, autonomous, integrated physical (e.g. robot) systems. The project especially aims at the integration of results from differents subfields of AI and Cognitive Science in a single embodied agent. We therefore investigate requirements for architectures, for forms of representation, for perceptual mechanisms, for learning, planning, reasoning, motivation, action, and communication. The results of this investigation will provide the basis for a succession of increasingly ambitious working systems to test and demonstrate the ideas.

The work in our research group focuses on the Planning capabilities needed by intelligent robots for successfully and sensibly acting and interacting with humans in a shared environment, including communication planning, continuous planning and acting, planning with incomplete knowledge, and cooperative multiagent planning.

The aim of the EU-funded project CogX is to develop robots that are able to work in open ended, challenging environments, dealing with novelty, uncertainty and change. In such environments, a robot should be able to reflect on the limits of its own abilities and knowledge ("self-understanding"), and be able to extend them based on its experiences and goals ("self-extension"). The high level aim of CogX is to develop a unified theory of self-understanding and self-extension with a convincing instantiation and implementation of this theory in a robot.

Work in our research group focuses on representations and methods enabling the robot to plan its physical actions as well as its interactions with humans and its internal processes. Furthermore, we want to enable the robot to decide which specic planning technique to use for a given task, which information to use for planning and what to learn from previous planning episodes.

The field of Computer Vision has brought forth many successful artificial vision systems, which are routinely and successfully used in a range of applications. However, to the extent that such systems embody knowledge and reasoning, it is "wired-in" or "compiled", in the sense that it is built into the code of the program, or encoded into a data structure. With such dedicated implementations it is hard to adapt (or generalise) the same basic image processing components to work in another domain, where different knowledge is required.

Aiming at a higher level of generality, the CogViSys project aims at building a vision system that is reusable by introducing self-adaptation at the level of perception and by making the knowledge base explicit at the level of reasoning, thus enabling the knowledge base to be changed. In order to make these ideas concrete, CogViSys researchers develop virtual commentators, who are able to translate visual information into textual descriptions. This is the unifying theme of the project, which is applied to three real-world scenarios:

1. traffic scene surveillance,

2. sign language interpretation, and

3. automated video annotation.

Within CogViSys, our research group focuses on the first application area, traffic scene surveillance.

Kontiplan, a joint research project with the German company Astrium GmbH, is mainly funded by the German Aerospace Center (Deutschen Zentrum für Luft- und Raumfahrt). The main goal of the project is to apply planning techniques in space scenarios like landing during interplanetary missions, servicing, or earth observation scenarios. In particular in time critiacal pahases of deep space missions in which there is no permanent communication link between the spacecraft and the ground station, these techniques are expected to increase robustness of the spacecraft towards uncertainty und unexpected events. By porting the developed algorithms to a real time operating system, the spacecraft is enabled to perform complex decicion making processes autonomously.

There are two main scenarios in the project: In a landing scenario, secure landing positions have to be reached while taking care of the dynamics, ressources, and possible actions the lander can actually execute. The other scenario is an earth observation scenario where the planner is used to for mission planning. The goal in this scenario is to observe certain areas on the ground while taking care of the actual wheather that might in render immpossible to observe some of the areas. For this purpose, the wheather behavior is modelled in a probablistic setting and probabilistic planning techniques are used to find good policies.

Robot assistants and service robots have not fully succeeded commercially yet. This is mainly due to the fact that even "simple" tasks of everyday life demand complex capabilities and high hard- and software reliability. In the German Service Robotics Initiave DESIRE, funded by BMBF, major German robotics research institutions and companies both study methodological foundations of service robotics for everyday life and also develop a reference platform, i.e. a modular robotic system for everyday applications.

Within DESIRE our research group studies how knowledge representation (ontologies) and action planning can increase flexibibility, reliability and usability of service robots.

The aim of this DFG project is to establish a direct cooperation between researchers in Action Formalisms, Description Logic and Planning so that they may reap the benefits of the advances both in theory and practice in each of the respective areas. The overall goal is to achieve a tight integration of systems from each area.

The work in our research group focuses on the subproject Planning Techniques and Action Languages. The recent developments in these areas shall be integrated to achieve expressive and efficient systems. The aim is to obtain comparability between Action Languages like GOLOG and FLUX on the one hand and Planning Languages like PDDL – which is regarded as standard in the field of planning – on the other hand. For this purpose a common semantic basis in the situation calculus shall be created and the expressive power shall be analyzed by means of compilation techniques.

In this project, we develop techniques for efficiently solving difficult planning problems by means of algorithms that test the satisfiability of propositional formulae. Topics that have been addressed and that will be addressed include symmetry reduction, partial-order reduction, different notions of parallel plans, finding optimal plans (with respect to a cost measure), finding suboptimal plans, and trade-offs between the quality and the cost of finding a plan, among others.

Planning is needed for deciding which actions to take in order to achieve given goals. It is needed by intelligent autonomous humans, animals, robots, and software agents for guaranteeing rational behavior in complex unpredictable environments. The research project addresses the planning problem faced in complex environments with nondeterminism and partial observability when the effects of actions cannot be unambiguously predicted and the environment can be incompletely observed.

In this project, we worked on the formal foundations of qualtitative spatial representations and the development of efficient algorithms for spatial reasoning. This included the following topics:

• Furthering the design and analysis of formal sematics for spatial calculi and increasing their expressiveness.

• Investigating soundness, completeness, and computational complexity properties of spatial calculi.

• Designing, implementing, and evaluating efficient inference methods for spatial calculi.

• Applying theoretical results from this project to empirical studies in cognitive psychology and using psychological results for reasoning algorithms, in cooperation with the MEMOSPACE DFG project.

The knowledge obtained in this project can be applied to diverse areas, including but not limited to natural language processing, document analysis, geographical information systems, and robot navigation.

RoboCup Soccer is an international initiative for supporting research in the areas of Artificial Intelligence and Autonomous Mobile Robots. Robotic soccer is used as a benchmark problem to directly compare results from the different research disciplines. RoboCup thus provides an opportunity for researchers to evaluate their approaches in direct competition with their international colleagues, thus furthering the state of the art in the respective research areas.

To play a good game of soccer, a team of robots must excel in a number of different problem areas, including sensing, orientation, moving in a dynamic environment, architecture of autonomous agents, cooperation in multi-agent systems, real-time processing, task planning, and machine learning. Experience in solving these problems and integrating them in a robot system can be applied to other contexts such as household robots, robots for the disabled, traffic navigation systems, and planetary rovers.

Our RoboCup soccer team CS Freiburg won the robotic soccer world championship (mid-size league) in 1998, 2000, and 2001.

KiRo is a completely autonomous table soccer playing robot: with the help of a camera it perceives the playing field and decides how the rods under its control should be moved depending on the current game situation.

KiRo was originally developed by Thilo Weigel within our research group, where it is utilized for basic research in the areas of Robotics and Artificial Intelligence. The system has been patented and is brought to market under the name StarKick by the Gauselmann Group.

Currently, KiRo is developed further by Dapeng Zhang, with the research focus on the application of learning techniques.

RobocupRescue Simulation is a recent project of the international RoboCup Association, which supports research on strategies for rescue missions in natural disaster areas. Rescue missions are conducted by so-called multi-agent systems, which are teams of individual programs, each of which is equipped with its own sensing, decision, and acting capabilities. The agents of a rescue team communicate with each other and act strategically to save as many lives, extinguish as many fires, and clear as many roads as possible. From a researcher's perspective, the scenario is interesting for providing generalizable knowledge about team work, communication and coordination techniques, action planning, real-time movement in dynamic environments, and the simulation of complex systems.

Our team ResQ Freiburg won the RoboCup simulation world championship in 2004.

The objective of this project is the development of real robots that can be used to find injured people in disaster areas. Different kinds of sensing techiques are applied for this purpose, e.g. CO2 sensors and infra-red cameras.

Our team Rescue Robots Freiburg won the second place at the GermanOpen 2005 in Paderborn.