Rewards in Action for Deep Reinforcement Learning of Industrial Robotics

Apr 25, 2022

ACROBA uses reward functions to evaluate and create neural networks for optimized perception-guided robot control in a wide range of industrial applications (such as deburring of flashes along the rims of container lids). These functions are designed with a specific robot, work-cell, parts to be grasped and assembled (motion and interaction task) and environment including human operators in mind. They are based on simple positive and negative distances between for instance robot and target object and objects to avoid, respectively. They form so to speak attractive and repulsive potential fields guiding the robot in acquiring a needed skill and performing accurately a certain task.

Want to know more ?

👉 Read the full article by Mr. NeC B.V.

Mr. NeC B.V., a high-tech SME based in Rotterdam, is involved in the ACROBA project and contributes to the development of deep (reinforcement) learning neural network architectures, models and algorithms for efficient plantwide human-robot collaboration.