Combining active learning and reactive control for robot grasping

By Oliver Kroemer, Renaud Detry, Justus H. Piater, Jan Peters. 2010

173 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Biomedical Engineering, Control and Systems Engineering, Meta & Continual Learning, Robotics & Control

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Biomedical Engineering

In Control and Systems Engineering

In Meta & Continual Learning

In Robotics & Control

Citations in the corpus

Learning Parameterized Skills (2012, methodology): “This upper bound is related to the one used to compute Upper Confidence Bound policies [41].”
Reinforcement learning in robotics: A survey (2013, background): “Using upper confidence bounds (Kaelbling, 1990) to direct exploration into regions with potentially high rewards, grasping can be learned efficiently (Kroemer et al., 2010).”
Learning robot in-hand manipulation with tactile features (2015, background): “In this section, we discuss related work on in-hand manipulation with and without tactile sensing, as well as reinforcement learning for various kinds of object manipulation.”
Active Reward Learning (2014, background): “Despite the variety of grasp stability measures that has been developed [34], it has been shown that the resulting grasps are outperformed by grasps learned from kinesthetic teachin [3, 23].”
Data-Driven Grasp Synthesis—A Survey (2013, methodology): “Several of the aforementioned grasp generation meth-ods [48, 38, 39] use the probabilistic approach towards object representation and pose estimation proposed by Detry et al. [55].”
Meta Reinforcement Learning with Latent Variable Gaussian Processes (2018, background): “In a robotics context, solutions for multiple related tasks are often desired, e.g., for grasping multiple objects [22] or in robot games, such as robot table tennis [24] or soccer [3].”
One-shot learning and generation of dexterous grasps for novel objects (2016, unknown): “In addition to these limitations, FA”
Multi-task policy search for robotics (2014, background): “, for grasping multiple objects [17] or in robot games, such as robot table tennis [18] or soccer [7].”
Learning motion primitive goals for robust manipulation (2011, background): “[9] also determine low-dimensional meta-parameters that automatically adapt the shape of the movement to a pre-specified goal, and apply it to grasping.”
Active learning of visual descriptors for grasping using non-parametric smoothed beta distributions (2012, methodology): “Also, we are currently studying connections with multi-arm bandit algorithms for planning optimal exploration strategies as used in [28], but exploiting the beta–binomial model.”
Acquiring visual servoing reaching and grasping skills using neural reinforcement learning (2013, methodology): “However, they also make extracting an accurate 3D pose of an object, which is used in most manipulation systems [1], difficult, and generally require a known system model as well as a calibrated camera for use in planning.”
Sample and Feedback Efficient Hierarchical Reinforcement Learning from Human Preferences (2018, background): “[25] propose a hierarchical architecture comprising an upper-level reinforcement learner for predicting the grasp location and a low-level reactive controller that generates the grasp motion.”
A Review of Robot Learning for Manipulation: Challenges, Representations, and Algorithms (2019, methodology): “As these parameters are often lower dimensional than full skills, active learning and multi-armed bandit methods can be used to select suitable argument values [267, 255, 221].”
Interactive Perception: Leveraging Action in Perception and Perception in Action (2016, methodology): “For the first three applications (Object Segmentation, Articulation Model Estimation and Object Dynamics Learning), we use a couple of simple examples (Figures 5 and 6) to allow the reader to better appreciate the benefits of IP and understand its distinction to Active Perception.”
Dex-Net 1.0: A cloud-based network of 3D objects for robust grasp planning using a Multi-Armed Bandit model with correlated rewards (2016, methodology): “We plan to study other quality metrics such as success on physical trials [25], [31] and alternate MAB methods based on upper confidence bounds [25], [32] or Gittins indices [26] in future work.”