LAB20 - HaptXDeep

Robotic systems for innovative manufacturing must be able to adapt quickly and flexibly to any changes in production requirements, product designs, materials, and suppliers, without the need for reprogramming. The advent of robot learning broadens the scope of potential applications that were previously beyond the capabilities of conventional automation techniques.

However, robots face the challenge of shortening the extensive learning phase while ensuring safety and reliability in industrial environments. Hence, enabling robots to learn human-like skills using innovative technology can accelerate adaptability to new manufacturing tasks and production processes, as well as ensure safety.

Aim

The aim of this project is to develop a robotics grasping learning system that can learn human-like skills. The rapid development of new technologies led to the continuous development of a new generation of industrial products and tools. Hence, adapting the robot to handle new objects more efficiently can be done by transferring human-like skills into its learning process. Additionally, a key focus of the project is to rigorously assess the safety and reliability of these adaptable robots. Proactive failure prediction methods will be used to minimize the risk of hazardous events.

Approach

• A 6-axis cobot with a robotic dexterous hand serves as a demonstrator.

• Implementation of imitation learning and a deep reinforcement learning system for robotic grasping.

• Digital twin-based error injection, deep learning-based anomaly detection, and AI-based failure prediction.

Benefit

Changeover time to new components is reduced as the robot adapts autonomously or with minimal human assistance. Remote guidance of the robot enables it to perform tasks in safety-critical areas.