2019 · 12 citations · 30 references
EngineeringRobotics EngineeringGraspingMarkov Decision ProcessesRoboticsFriction ModelingGrasp PlanningGrasp Pose EstimationGrasp Quality MetricsPoint Cloud ProcessingFrictionGrasp ReliabilityGrasp Robustness Under UncertaintyGrasp Uncertainty EstimationMarkov Decision ModelsMonte Carlo SamplingRobust Grasp PlanningRobust Grasping
When robust vacuum suction grasps are not accessible, toppling can change an object's 3D pose to provide access to suction grasps. We extend prior toppling models by characterizing the toppling reliability for a 3D object specified by a triangular mesh, using Monte-Carlo sampling to model uncertainty in pose, friction coefficients, and push direction. The model estimates the resulting distribution of object poses following a topple action. We generate a dataset of toppling analysis for 1,257,000 candidate points on the surface of 189 3D meshes and perform 700 physical toppling experiments using an ABB YuMi. We find that the model outperforms a Max-Height baseline model by a percent difference of 21.3% when comparing the total variation distance between each model's predicted probability distribution against the empirical distribution. We use the proposed model as the state transition function in a Markov Decision Process (MDP) to plan optimal sequences of toppling actions to expose access to robust suction grasps. Data from 20,000 simulated rollouts suggest the proposed Value Iteration Policy can increase suction grasp reliability by 33.6%, computed using grasp analysis from Dexterity Network (Dex-Net) 3.0. Code, datasets, and videos can be found at https://sites.google.com/view/toppling.
30
Reinforcement Learning: An Introduction
IEEE Transactions on Neural Networks · 2005 · 25.8K citations
Orienting polygonal parts without sensors
Ken Goldberg · Algorithmica · 1993 · 444 citations