In recent years, there has been a growing interest in designing multi-robot systems to provide cost-effective, fault-tolerant and reliable solutions to a variety of automated applications. In particular, from an industrial perspective, cooperative carrying techniques based on Reinforcement Learning (RL) gained a strong interest. Compared to a single robot system, this approach improves the system’s robustness and manipulation dexterity in the transportation of large objects. However, in the current state of the art, the environments’ dynamism and re-training procedure represent a considerable limitation for most of the existing cooperative carrying RL-based solutions. In this thesis, we employ the Value Propagation Networks (VPN) algorithm for cooperative multi-robot transport scenarios. We extend and test the Delta-Q cooperation metric to V-value-based agents, and we investigate path generation algorithms and trajectory tracking controllers for differential drive robots. Moreover, we explore localization algorithms in order to take advantage of range sensors and mitigate the drift errors of wheel odometry, and we conduct experiments to derive key performance indicators of range sensors' precision. Lastly, we perform realistic industrial indoor simulations using Robot Operating System (ROS) and Gazebo 3D visualization tool, including physical objects and 6G communication constraints. Our results showed that the proposed VPN-based algorithm outperforms the current state-of-the-art since the trajectory planning and dynamic obstacle avoidance are performed in real-time, without re-training the model, and under constant 6G network coverage.
In recent years, there has been a growing interest in designing multi-robot systems to provide cost-effective, fault-tolerant and reliable solutions to a variety of automated applications. In particular, from an industrial perspective, cooperative carrying techniques based on Reinforcement Learning (RL) gained a strong interest. Compared to a single robot system, this approach improves the system’s robustness and manipulation dexterity in the transportation of large objects. However, in the current state of the art, the environments’ dynamism and re-training procedure represent a considerable limitation for most of the existing cooperative carrying RL-based solutions. In this thesis, we employ the Value Propagation Networks (VPN) algorithm for cooperative multi-robot transport scenarios. We extend and test the Delta-Q cooperation metric to V-value-based agents, and we investigate path generation algorithms and trajectory tracking controllers for differential drive robots. Moreover, we explore localization algorithms in order to take advantage of range sensors and mitigate the drift errors of wheel odometry, and we conduct experiments to derive key performance indicators of range sensors' precision. Lastly, we perform realistic industrial indoor simulations using Robot Operating System (ROS) and Gazebo 3D visualization tool, including physical objects and 6G communication constraints. Our results showed that the proposed VPN-based algorithm outperforms the current state-of-the-art since the trajectory planning and dynamic obstacle avoidance are performed in real-time, without re-training the model, and under constant 6G network coverage.
Cooperative Carrying Control for Mobile Robots in Indoor Scenario
CANEVARO, ALESSANDRO
2022/2023
Abstract
In recent years, there has been a growing interest in designing multi-robot systems to provide cost-effective, fault-tolerant and reliable solutions to a variety of automated applications. In particular, from an industrial perspective, cooperative carrying techniques based on Reinforcement Learning (RL) gained a strong interest. Compared to a single robot system, this approach improves the system’s robustness and manipulation dexterity in the transportation of large objects. However, in the current state of the art, the environments’ dynamism and re-training procedure represent a considerable limitation for most of the existing cooperative carrying RL-based solutions. In this thesis, we employ the Value Propagation Networks (VPN) algorithm for cooperative multi-robot transport scenarios. We extend and test the Delta-Q cooperation metric to V-value-based agents, and we investigate path generation algorithms and trajectory tracking controllers for differential drive robots. Moreover, we explore localization algorithms in order to take advantage of range sensors and mitigate the drift errors of wheel odometry, and we conduct experiments to derive key performance indicators of range sensors' precision. Lastly, we perform realistic industrial indoor simulations using Robot Operating System (ROS) and Gazebo 3D visualization tool, including physical objects and 6G communication constraints. Our results showed that the proposed VPN-based algorithm outperforms the current state-of-the-art since the trajectory planning and dynamic obstacle avoidance are performed in real-time, without re-training the model, and under constant 6G network coverage.File | Dimensione | Formato | |
---|---|---|---|
MasterThesis_ACanevaro.pdf
accesso aperto
Dimensione
2.78 MB
Formato
Adobe PDF
|
2.78 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/48209