A framework for the integration of deep learning-based systems in humanoid robots

Robots are typically limited by their processing power, memory, and storage capacity compared to modern computers. This limitation can restrict the complexity of customized applications that can be integrated into them to enhance their social or active interaction skills. To address this challenge, we propose a framework with a client-server-like architecture. The framework enables energy-hungry AI applications to be processed on the cloud-server side, while humanoid robots on the client side only acquire data from their external sensors and use the output from the server to continue with interactions. We present a case study using Softbank's humanoid robot NAO to evaluate the effectiveness of this solution. The proposed framework has the potential to enhance the social or active interaction skills of humanoid robots by enabling more complex customized applications to be integrated into them.