1309.2660 - hassony2/inria-research-wiki GitHub Wiki

2013 IEEE Transactions on Robotics

[1309.2660] Data-Driven Grasp Synthesis - A Survey [PDF] [notes]

Jeanette Bohg, Antonio Morales, Tamin Asfour, Danica Kragic

read 22/05/2017

Synthesis

Analytic approach

force-closure grasps

Analytic methods try to construct force-closure grasps which have the following properties :

  • dexterous
  • in equilibrium
  • stable
  • exhibit a certain dynamic behaviour

Challenges :

  • simplified models/approximation: Coulomb friction/rigid-body modelling

Caging

For 'realistic' grasps, caging implies that the object can not move to infinity in any direction from the given grasp

Limits

Poor transfer from simulation to reality : analytical solutions computed in simulations sometimes do not transfer to reality given approximation errors in the modelling process

Stability (what happens when small perturbation is applied) is not taken into account

Data driven

Learning discriminative object features to infer suitable grasp configuration

Possibility to use primitives to achieve good but not optimal grasps