A5050112206- Formal Methods in Verification
- Logic, programming, and type systems
- Software Testing and Debugging Techniques
- Petri Nets in System Modeling
- Embedded Systems Design Techniques
- Model-Driven Software Engineering Techniques
- Advanced Software Engineering Methodologies
- Adversarial Robustness in Machine Learning
- Machine Learning and Algorithms
- Real-Time Systems Scheduling
- Software Reliability and Analysis Research
- Software Engineering Research
- Distributed systems and fault tolerance
- semigroups and automata theory
- Logic, Reasoning, and Knowledge
- Reinforcement Learning in Robotics
- Safety Systems Engineering in Autonomy
- Radiation Effects in Electronics
- Scheduling and Optimization Algorithms
- VLSI and Analog Circuit Testing
- Advanced Multi-Objective Optimization Algorithms
- Smart Grid Security and Resilience
- Autonomous Vehicle Technology and Safety
- Experimental Behavioral Economics Studies
- Evolutionary Game Theory and Cooperation
Clausthal University of Technology
2019-2024
University of California, Irvine
2022
Aalborg University
2022
RWTH Aachen University
2022
Toyota Industries (United States)
2022
Amazon (United States)
2022
University of Manchester
2022
California University of Pennsylvania
2022
University of Milano-Bicocca
2022
Langley Research Center
2022
Reinforcement learning algorithms discover policies that maximize reward, but do not necessarily guarantee safety during or execution phases. We introduce a new approach to learn optimal while enforcing properties expressed in temporal logic. To this end, given the logic specification is be obeyed by system, we propose synthesize reactive system called shield. The shield monitors actions from learner and corrects them only if chosen action causes violation of specification. discuss which...
We consider a problem on the synthesis of optimal reactive controllers with an priori unknown performance criterion while satisfying given temporal logic specification through interaction uncontrolled environment. decouple into two sub-problems. First, we extract (maximally) permissive strategy for system, which encodes multiple (possibly all) ways in system can react to adversarial environment and satisfy specifications. Then, quantify as (still unknown) reward function, compute - by using...
As robots begin to interact closely with humans, we need build systems worthy of trust regarding the safety and quality interaction.
We consider the synthesis of reactive systems that are robust against intermittent violations their environment assumptions. Such assumptions needed to allow many work in a larger context fulfill tasks. Yet, due glitches hardware or exceptional operating conditions, these do not always hold field. Manually constructed often exhibit error-resilience and can continue correctly such cases. With development cycles becoming shorter, thus an increasingly suitable alternative manual design systems,...
The increased interest in reactive synthesis over the last decade has led to many improved solutions but also new questions. In this paper, we discuss question of how deal with assumptions on environment behavior. We present four goals that think should be met and review several different possibilities have been proposed. argue each them falls short at least one aspect.
Multi-agent reinforcement learning (MARL) has been increasingly used in a wide range of safety-critical applications, which require guaranteed safety (e.g., no unsafe states are ever visited) during the process.Unfortunately, current MARL methods do not have guarantees. Therefore, we present two shielding approaches for safe MARL. In centralized shielding, synthesize single shield to monitor all agents' joint actions and correct any action if necessary. factored multiple shields based on...
Synthesis of correct-by-construction robot controllers from high-level specifications has the advantage providing guaranteed behavior under different environments.Typically, when such are synthesized, assumptions that user makes about environment, if any, incorporated into resulting controller.In practice, however, environment may be unknown to user, thus preventing application synthesis.Even available, they not hold during robot's execution due modeling errors or unforeseen anomalous...
By performing synthesis from formal high-level mission specifications, we can obtain robot controllers that are guaranteed to operate correctly under the specified environment conditions. Such conditions must be stated in specification whenever there is no way which robot's task fulfilled without them holding, and they relate possible behaviors of with behavior robot. Contemporary algorithms however frequently construct implementations try trivially satisfy their specifications by actively...
Abstract Today, arc-based additive manufacturing has great potential for industrial application due to new developments in robotics, welding technology, and computer-aided manufacturing. Two issues are currently the focus of research. One is accurate generation geometry with respect design, e.g., fidelity, defined roughness, shape deviations within tolerances. Here, there still open questions, particularly regard path planning dependence on selected process variables. The second topic...
We present constraint matrix diagrams (CMDs), a novel data structure for the fully symbolic reach ability analysis of timed automata. CMDs combine matrix-based and diagram-based state space representations generalizing concepts difference bound matrices (DBMs), clock (CDDs), restriction (CRDs). The key idea is to represent convex parts as (partial) DBMs which are, in turn, organized CDD/CRD-like ordered reduced diagram. location information incorporated special Boolean matrices. describe all...