A5068140202- Robot Manipulation and Learning
- Reinforcement Learning in Robotics
- Robotic Path Planning Algorithms
- Adversarial Robustness in Machine Learning
- Cancer Diagnosis and Treatment
- Robotic Locomotion and Control
- Modular Robots and Swarm Intelligence
- Advanced Vision and Imaging
- Multimodal Machine Learning Applications
- Domain Adaptation and Few-Shot Learning
- Head and Neck Cancer Studies
- Formal Methods in Verification
- Image Processing Techniques and Applications
- AI-based Problem Solving and Planning
- Brain Metastases and Treatment
- Head and Neck Surgical Oncology
- Reconstructive Surgery and Microvascular Techniques
University of Leeds
2018-2021
Deutsches Zentrum für Luft- und Raumfahrt e. V. (DLR)
2016
We address the manipulation task of retrieving a target object from cluttered shelf. When is hidden, robot must search through clutter for it. Solving this requires reasoning over likely locations object. It also physics multi-object interactions and future occlusions. In work, we present data-driven hybrid planner generating occlusion-aware actions in closed-loop. The explores occluded as predicted by learned distribution observation stream. guided heuristic trained with reinforcement...
Manipulation in clutter requires solving complex sequential decision making problems an environment rich with physical interactions. The transfer of motion planning solutions from simulation to the real world, open-loop, suffers inherent uncertainty modelling world physics. We propose interleaving and execution real-time, a closed-loop setting, using Receding Horizon Planner (RHP)for pushing manipulation clutter. In this context, we address problem finding suitable value function based...
Universal robotic agents are envisaged to perform a wide range of manipulation tasks in everyday environments. A common action observed many household chores is wiping, such as the absorption spilled water with sponge, skimming breadcrumbs off dining table, or collecting shards broken mug using broom. To cope this versatility, have represent on high level abstraction. In work, we propose medium wiping (e.g. water, breadcrumbs, shards) generic particle distribution. This representation...
Physics-based manipulation in clutter involves complex interaction between multiple objects. In this paper, we consider the problem of learning, from a physics simulator, skills to solve multi-step sequential decision making real world. Our approach has two key properties: (i) ability generalize and transfer (over type, shape, number objects scene) using an abstract image-based representation that enables neural network learn useful features; (ii) perform look-ahead planning image space...
Manipulation in clutter requires solving complex sequential decision making problems an environment rich with physical interactions. The transfer of motion planning solutions from simulation to the real world, open-loop, suffers inherent uncertainty modelling world physics. We propose interleaving and execution real-time, a closed-loop setting, using Receding Horizon Planner (RHP) for pushing manipulation clutter. In this context, we address problem finding suitable value function based...
Physics-based manipulation in clutter involves complex interaction between multiple objects. In this paper, we consider the problem of learning, from a physics simulator, skills to solve multi-step sequential decision making real world. Our approach has two key properties: (i) ability generalize and transfer (over type, shape, number objects scene) using an abstract image-based representation that enables neural network learn useful features; (ii) perform look-ahead planning image space...
We address the manipulation task of retrieving a target object from cluttered shelf. When is hidden, robot must search through clutter for it. Solving this requires reasoning over likely locations object. It also physics multi-object interactions and future occlusions. In work, we present data-driven hybrid planner generating occlusion-aware actions in closed-loop. The explores occluded as predicted by learned distribution observation stream. guided heuristic trained with reinforcement...