Integrating convolutional neural networks and reinforcement learning for autonomous person-following in social robotics

The integration of advanced robotics with sophisticated machine learning techniques offers substantial opportunities to enhance the autonomous capabilities of social robots. This paper presents a novel approach that enables a Pepper-type social robot to autonomously follow a person. Our methodology...

Full description

Autores:
Ortiz Almanza, David Santiago
Tipo de recurso:
Trabajo de grado de pregrado
Fecha de publicación:
2024
Institución:
Universidad de los Andes
Repositorio:
Séneca: repositorio Uniandes
Idioma:
eng
OAI Identifier:
oai:repositorio.uniandes.edu.co:1992/74468
Acceso en línea:
https://hdl.handle.net/1992/74468
Palabra clave:
Person following
Convolutional neural networks
Reinforcement learning
Social robotics
Pepper robot
Ingeniería
Rights
openAccess
License
Attribution-NonCommercial 4.0 International
Description
Summary:The integration of advanced robotics with sophisticated machine learning techniques offers substantial opportunities to enhance the autonomous capabilities of social robots. This paper presents a novel approach that enables a Pepper-type social robot to autonomously follow a person. Our methodology combines Convolutional Neural Networks (CNNs) and Reinforcement Learning (RL) within a modular architecture. This architecture utilizes both 2D and 3D cameras as sensors, which provide robust real-time detection and tracking, while RL allows adaptive movement control. We have tested our implementation, demonstrating high performance and reliability across various complex environments. The results from these tests confirm that the proposed architecture and learning methods greatly enhance the robot’s ability to autonomously follow a person. This research represents a significant advancement in the practical deployment of autonomous systems in the field of social robotics.