Optimization of Robot Grasping Using Reinforcement Learning in Simulation

Springer eBooks, pp. 305–316

Abstract

Abstract This paper presents a novel approach to optimizing robot grasping through reinforcement learning in simulation, emphasizing the integration of tactile feedback alongside visual cues to enhance object grasping efficiency. Traditional methods primarily depend on visual feedback, often falling short in achieving optimal grasping. Our methodology proposes a cost-effective solution that significantly improves the software components responsible for object grasping. We introduce a two-phase process: a pre-grasp phase focusing on positioning and a grasping phase that adjusts based on tactile inputs. This approach demonstrates improved grasping outcomes, particularly in environments with high noise levels, underscoring the potential of tactile information in robotic grasping optimization.

Authors 4

1. Manuel Belke corresponding Aachen Laboratory for Machine Tools and Production Engineering (WZL)

   RWTH Aachen University

   Affiliation as printed

   Laboratory for Machine Tools and Production Engineering (WZL) of RWTH Aachen University, Aachen, Germany

2. Madhura Phatak Aachen Laboratory for Machine Tools and Production Engineering (WZL)

   RWTH Aachen University

   Affiliation as printed

   Laboratory for Machine Tools and Production Engineering (WZL) of RWTH Aachen University, Aachen, Germany

3. Oliver Petrović Aachen Laboratory for Machine Tools and Production Engineering (WZL)

   RWTH Aachen University

   Affiliation as printed

   Laboratory for Machine Tools and Production Engineering (WZL) of RWTH Aachen University, Aachen, Germany

4. Christian Brecher Aachen Laboratory for Machine Tools and Production Engineering (WZL)

   RWTH Aachen University

   Affiliation as printed

   Laboratory for Machine Tools and Production Engineering (WZL) of RWTH Aachen University, Aachen, Germany