Researchers at TU Delft are working towards a future in which a single AI system can control multiple types of robots. Daan Brinks (ImPhys), together with Laurens Engwegen and Wendelin Böhmer from EWI, are developing an AI agent that can transfer knowledge between different robotic platforms, reducing the need to train a separate AI system for every individual robot.
Traditionally, AI agents are trained to operate one specific robot and perform the various tasks assigned to that machine. The researchers’ new approach enables an AI agent to reuse knowledge gained from one robot when controlling another. This could significantly accelerate the deployment of robots in new environments and applications while reducing the amount of training data and time required.
By creating an AI agent that can adapt across different robotic systems, the team aims to make robotics more flexible, scalable, and efficient. Such advances could help bring autonomous systems closer to real-world applications where robots must operate in diverse and changing conditions.
The significance of this work has been recognized with the acceptance of the team’s paper at NeurIPS (Conference on Neural Information Processing Systems), widely regarded as the world’s leading conference in artificial intelligence and machine learning.