Leveraging big data for grasp planning

By Daniel Kappler, Jeannette Bohg, Stefan Schaal. 2015

307 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Control and Systems Engineering

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 Control and Systems Engineering

Citations in the corpus

Learning hand-eye coordination for robotic grasping with deep learning and large-scale data collection (2016, methodology): “This is more than an order of magnitude larger than prior methods based on direct self-supervision (Pinto and Gupta, 2016) and more than double the dataset size of prior methods based on synthetic grasps from 3D scans (Kappler et al., 2015).”
Supersizing self-supervision: Learning to grasp from 50K tries and 700 robot hours (2015, methodology): “For gripping we use the stock two ﬁngered parallel gripper with a maximum width (open state) of 75 mm and a minimum width (close state) of 37 mm.”
Meta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement Learning (2019, background): “Similarly, there exists a number of robotics simulation benchmarks including visual navigation [39, 40, 41, 42, 43], autonomous driving [44, 45, 46], grasping [47, 48, 49], single-task manipulation [50], among others.”
Using Simulation and Domain Adaptation to Improve Efficiency of Deep Robotic Grasping (2018, background): “A number of methods combine both ideas, for example using analytic metrics to label training data [3], [7].”
Supervision via competition: Robot adversaries for learning tasks (2016, background): “Significant progress has been made recently using data-driven solutions for training deep learned models on both corporeal robots [1]–[4] and in simulation [34], [35].”
Sim-To-Real via Sim-To-Sim: Data-Efficient Robotic Grasping via Randomized-To-Canonical Adaptation Networks (2018, methodology): “These methods commonly make use of either hand-labeled grasp positions [33, 28], self-supervision [44], or predicting grasp outcomes [34].”
Deep Reinforcement Learning for Vision-Based Robotic Grasping: A Simulated Comparative Evaluation of Off-Policy Methods (2018, background): “Deep learning has been recently incorporated into such systems [20], [22], [24], [26], [32].”
Multi-Task Domain Adaptation for Deep Learning of Instance Grasping from Simulation (2017, unknown): “With recent progress in deep learning and reinforcement learning, large scale learning-based methods have been used in robotic manipulation [1], [2], [3].”
Dex-Net 3.0: Computing Robust Robot Suction Grasp Targets in Point Clouds using a New Analytic Model and Deep Learning (2017, methodology): “A popular approach is to use human labels of graspable regions in RGB-D images [19] or point clouds [5], [15] to learn a grasp detector with computer vision techniques.”
Learning a visuomotor controller for real world robotic grasping using simulated depth images (2017, background): “This is a general advantage of a closed-loop controller relative to typical grasp perception methods [4], [8], [3], [7], [2].”
Learning ambidextrous robot grasping policies (2019, background): “Training datasets are collected from humans ( 7 – 9 ) or physical experiments ( 10 – 12 ).”
Deep learning in robotics: a review of recent research (2017, background): “Kappler, Bohg, and Schaal [133] generated a database of over 300,000 grasps on over 700 objects in simulation, generating physics-based grasp quality metrics for each and using this to classify grasp stability automatically.”
Deep learning a grasp function for grasping under gripper pose uncertainty (2016, background): “In [11], [15], [16], it has been shown that dynamic physics simulations offer a more accurate prediction of grasp quality than the standard static metrics.”
Action Image Representation: Learning Scalable Deep Grasping Policies with Zero Real World Data (2020, methodology): “Since manual annotation can be timeconsuming, simulation approaches have been exploited to generate real-world like sensory data [13], [17], [24], [25].”
Dex-Net 3.0: Computing Robust Vacuum Suction Grasp Targets in Point Clouds Using a New Analytic Model and Deep Learning (2018, methodology): “A common approach is to select a configuration that maximizes a quality metric (or reward) based on wrench space analysis [24], robustness to perturbations [32], or a model learned from human labels [14] or self-supervision [25].”
Grasp'D: Differentiable Contact-rich Grasp Synthesis for Multi-fingered Hands (2022, methodology): “While they are more computationally costly, simulation-based metrics for parallel-jaw grasps better align with human judgement [48] and with real world performance [17, 24,”
Grasp Pose Detection in Point Clouds (2017, methodology): “mpts) between 75% and 95% for novel objects presented in isolation or in light clutter (Fischinger and Vincze 2012; Fischinger et al. 2013; Detry et al. 2013; Herzog et al. 2012; Kroemer et al. 2012; Kappler et al. 2015; ten Pas and Platt 2015). Not only are these success rates too low for practical”
Learning Hand-Eye Coordination for Robotic Grasping with Large-Scale Data Collection (2016, methodology): “We trained our method using over 800,000 grasp attempts on a very large variety of objects, which is more than an order of magnitude larger than prior methods based on direct self-supervision [Pinto and Gupta, 2016] and more than double the dataset size of prior methods based on synthetic grasps fro”
COG: Connecting New Skills to Past Experience with Offline Reinforcement Learning (2020, background): “However, most prior work in this category focuses on executing the same actions on novel objects [24, 25, 26, 27, 28].”