
GPAtlasRRT: An exploration strategy for novel object shape modeling
http://www.worldscientific.com/doi/abs/10.1142/S0219843618500147
In this work, we use a Gaussian Process (GP) as such representation. Then, using the fact that the 0-levelset of the GP - the surface of the object - is an implicitly dened manifold, we borrow the AtlasRRT algorithm concept to simultaneously: (i) build an atlas via continuation methods that locally parameterizes the object and that is used to select the next-best touch, and (ii) use an RRT-like strategy to dev

Feature-Based Transfer Learning for Robotic Push Manipulation
This work presents a data-efficient approach to learning transferable forward models for robotic push manipulation. Our approach extends previous work on modular contact-based predictors by leveraging information on the pushed object's local surface features. We test the hypothesis that, by conditioning predictions on local surface features, we can achieve generalisation across objects of different shapes. Our approach involves learning motion models that are specific to con
Grasping a Shape with Uncertain Location
Successful grasp planning requires an appropriate finger placement for which object geometry and location need to be known. Here we investigate how position uncertainty and shape influence the selection of a two-finger pinch grasp. Elliptical cylinders were stereoscopically presented in rapid succession. The position of each cylinder was randomly selected using two orthogonal Gaussian distribution whose orientation changed at each trial. The axes of the elliptical base were a

Two-level RRT Planning for Robotic Push Manipulation
Pushing operations are encountered frequently in robotics, but have received comparatively little attention in the research comunity. In one sense, pushing is perhaps the most primitive kind of manipulation, but the relationship between applied pushes and the resulting workpiece motions are complex and hard to predict and control. Push contacts are also important to more complex tasks such as grasping. This work presents an algorithm for planning sequences of pushes, by which


Exploratory Reach-to-Grasp Trajectories for Uncertain Object Poses
This work addresses the problem of planning the reach-to-grasp trajectory for a robotic arm and hand, when there is uncertainty in the pose of the object being grasped. If the object is not in its expected location, then the robot may still gain additional information about the object pose by making tactile or haptic observations if a finger or other part of the hand collides with part of the object during the reach-to-grasp operation. Therefore, it is desirable to plan the r


Sequential Trajectory Re-planning for Dexterous Grasping
This work, firstly, describes how to iteratively update localisation knowledge using tactile observations from a previous grasp attempt; secondly, shows how successive grasp trajectories can be planned with respect to these iteratively refined object poses; and, thirdly, shows how each reach-to-grasp trajectory can be deliberately planned to maximise new tactile information gain, while also reaching towards the expected grasp location derived from previous information. #Robot


Planning Trajectories under Object Pose Uncertainty
Planning in robotics means coping with dynamic and uncertain worlds. Unstructured worlds are perceived by the agent through noisy sensors as laser, camera and so on. Especially, when the agent works "in contact" with the environment, uncertainty plays a crucial role. In such cases, the agent interacts actively with the surrounding environment and its actions affect future states of the environment itself. In other words, future observations are affected by earlier agent's act