A5004137684- Robot Manipulation and Learning
- Formal Methods in Verification
- Reinforcement Learning in Robotics
- AI-based Problem Solving and Planning
- Robotic Path Planning Algorithms
- Logic, programming, and type systems
- Modular Robots and Swarm Intelligence
- Manufacturing Process and Optimization
- Robotics and Sensor-Based Localization
- Model-Driven Software Engineering Techniques
- Soft Robotics and Applications
- Advanced Neural Network Applications
- Logic, Reasoning, and Knowledge
- Advanced Software Engineering Methodologies
- Flexible and Reconfigurable Manufacturing Systems
- Robotic Locomotion and Control
- Distributed systems and fault tolerance
- Social Robot Interaction and HRI
- Image Processing Techniques and Applications
- Advanced Malware Detection Techniques
- Advanced Control Systems Optimization
- Evolutionary Algorithms and Applications
- Video Surveillance and Tracking Methods
- Human Pose and Action Recognition
- Multimodal Machine Learning Applications
Robert Bosch (Germany)
2016-2024
Robert Bosch (India)
2021-2023
Institute of Electrical and Electronics Engineers
2021
Gorgias Press (United States)
2021
Stevens Institute of Technology
2021
KTH Royal Institute of Technology
2017-2018
Technical University of Darmstadt
2013-2016
University of Stuttgart
2016
This paper describes a framework for automatically generating optimal action-level behavior team of robots based on temporal logic mission specifications under resource constraints. The proposed approach optimally allocates separable tasks to available robots, without requiring priori an explicit representation the or computation all task execution costs. Instead, we propose identifying sub-tasks in automaton specification and simultaneously allocating planning their execution. avoids need...
Motivated by the DARPA Robotics Challenge (DRC), application of operator assisted (semi-)autonomous robots with highly complex locomotion and manipulation abilities is considered for solving tasks in potentially unknown unstructured environments. Because limited a priori knowledge about state environment needed to achieve mission, sufficiently complete design high level robot behaviors not possible. Most situational required such behavior gathered only during runtime needs be interpreted...
In this work, we take a step towards bridging the gap between theory of formal synthesis and its application to real-world, complex, robotic systems. particular, present an end-to-end approach for automatic generation code that implements high-level robot behaviors in verifiably correct manner, including reaction possible failures low-level actions. We start with description system defined priori. Thus, non-expert user need only specify task. automatically construct specification, fragment...
Team ViGIR and Hector participated in the DARPA Robotics Challenge (DRC) Finals, held June 2015 Pomona, California, along with 21 other teams from around world. Both competed using same high‐level software, conjunction independently developed low‐level software specific to their humanoid robots. On basis of previous work on operator‐centric manipulation control at level affordances, we an approach that allows one or more human operators share authority a behavior controller. This...
Sensing, planning, controlling, and reasoning, are human-like capabilities that can be artificially replicated in an autonomous robot. Such a robot implements data structures algorithms devised on large spectrum of theories, from probability theory, mechanics, control theory to ethology, economy, cognitive sciences. Software plays key role the development robotic systems, as it is medium embody intelligence machine. During last years, however, software increasingly becoming bottleneck...
Enabling robots to quickly learn manipulation skills is an important, yet challenging problem. Such should be flexible, e.g., able adapt the current workspace configuration. Furthermore, accomplish complex tasks, sequence several and them changing situations. In this work, we propose a rapid robot skill-sequencing algorithm, where are encoded by object-centric hidden semi-Markov models. The learned skill models can encode multimodal (temporal spatial) trajectory distributions. This approach...
We present an efficient approach to plan action sequences for a team of robots from single finite LTL mission specification. The resulting execution strategy is proven solve the given with minimal costs, e.g., shortest time. For planning, established graph-based search method based on multi-objective path problem adapted multi-robot planning and extended support resource constraints. further improve efficiency significantly missions which consist independent parts by using previous results...
Coordinating a team of robots to fulfill common task is still demanding problem. This even more the case when considering uncertainty in environment, as well temporal dependencies within specification. A multi-robot cooperation from single goal specification requires mechanisms for decomposing an efficient planning team. However, action sequences offline insufficient real world applications. Rather, due uncertainties, also need closely coordinate during execution and adjust their policies...
This paper presents a novel method for model-free prediction of grasp poses suction grippers with multiple cups. Our approach is agnostic to the design gripper and does not require gripper-specific training data. In particular, we propose two-step approach, where first, neural network predicts pixel-wise quality an input image indicate areas that are generally graspable. Second, optimization step determines optimal selection corresponding based on configured layouts activation schemes....
While recent advances in approaches for control of humanoid robot systems show highly promising results, consideration fully integrated solving complex tasks such as disaster response has only gained focus recently. In this work, a software framework robots is introduced. It provides newcomers well experienced researchers robotics comprehensive system comprising open source packages locomotion, manipulation, perception, world modeling, behavior control, and operator interaction. The uses the...
For performing robotic manipulation tasks, the core problem is determining suitable trajectories that fulfill task requirements. Various approaches to compute such exist, being learning and optimization main driving techniques. Our work builds on learning-from-demonstration (LfD) paradigm, where an expert demonstrates motions, robot learns imitate them. However, demonstrations are not sufficient capture all sorts of specifications, as timing grasp object. In this paper, we propose a new...
Planning efficient and coordinated policies for a team of robots is computationally demanding problem, especially when the system faces uncertainty in outcome or duration actions.In practice, approximation methods are usually employed to plan reasonable an acceptable time.At same time, many typical robotic tasks include repetitive pattern.On one hand, this multiplies increased cost inefficient solutions.But on other it also provides potential improving initial, solution over time.In paper,...
In social and industrial facilities of the future such as hospitals, hotels, warehouses, teams robots will be deployed to assist humans in accomplishing everyday tasks like object handling, transportation, or pickup delivery operations. a context, different (e.g., mobile platforms, static manipulators, manipulators) with actuation, manipulation, perception capabilities must coordinated achieve various complex cooperative parts assembly automotive industry loading unloading palettes...
Dense Object Nets (DONs) by Florence, Manuelli and Tedrake (2018) introduced dense object descriptors as a novel visual representation for the robotics community. It is suitable many applications including grasping, policy learning, etc. DONs map an RGB image depicting into descriptor space image, which implicitly encodes key features of invariant to relative camera pose. Impressively, self-supervised training can be applied arbitrary objects evaluated deployed within hours. However,...
Hierarchical Finite State Machines (HFSM) are commonly used as high-level behavioral control strategies for robotic systems. Over the years a number of behavior engines have been developed original Robot Operating System (ROS 1), including popular Flexible Behavior Engine (FlexBE). In recent new ROS 2 system has to improve communication performance by using Data Distribution Service (DDS) protocol. This paper describes conversion both FlexBE and open-source Navigation latest release, our...
We propose a self-supervised training approach for learning view-invariant dense visual descriptors using image augmentations. Unlike existing works, which often require complex datasets, such as registered RGBD sequences, we train on an unordered set of RGB images. This allows from single camera view, e.g., in robotic cell with fix-mounted camera. create synthetic views and pixel correspondences data find our are competitive to the methods, despite simpler recording setup requirements. show...
This paper studies the problem of penalizing rule violation in context logic-based motion planning. Translating a given Linear Temporal Logic (LTL) into penalty structure requires design decision, since discrete automata obtained from do not provide straightforward method to penalize violation. We propose that explicitly specifies allow for more flexibility parametrization desired behaviors and differentiation semantics. Case study results are shown an autonomous driving scenario.
The prevailing grasp prediction methods predominantly rely on offline learning, overlooking the dynamic learning that occurs during real-time adaptation to novel picking scenarios. These scenarios may involve previously unseen objects, variations in camera perspectives, and bin configurations, among other factors. In this paper, we introduce a approach, SSL-ConvSAC, combines semi-supervised reinforcement for online learning. By treating pixels with reward feedback as labeled data others...