Tesi etd-02022024-142421
Link copiato negli appunti
Tipo di tesi
Dottorato
Autore
ASTOLFI, ANNA
URN
etd-02022024-142421
Titolo
Leveraging reduced-order models for the control of underwater and terrestrial legged robots
Settore scientifico disciplinare
ING-IND/34
Corso di studi
Istituto di Biorobotica - PHD IN BIOROBOTICA
Commissione
relatore Dott. CALISTI, MARCELLO
Presidente Prof. ODDO, CALOGERO MARIA
Membro MARIA ELENA GIANNACCINI
Presidente Prof. ODDO, CALOGERO MARIA
Membro MARIA ELENA GIANNACCINI
Parole chiave
- bioinspired control
- legged robots
- reduced-order model
- spring- loaded inverted pendulum
- underwater robotics
Data inizio appello
03/07/2024;
Disponibilità
parziale
Riassunto analitico
Legged robotics represents a fascinating area of research and development that draws inspiration from the locomotion abilities of animals pushing the boundaries of engineering but also deepening our understanding of locomotion principles in nature.
Despite the remarkable progress in the field, several challenges limit the widespread use of legged robotics in field operations, including locomotion control and stability, power efficiency, and terrain adaptability.
The work carried out in this thesis focused on addressing these limitations through mathematical modeling grounded on biomechanical studies. Significant emphasis was placed on adapting the capabilities of terrestrial legged systems to novel environments, such as underwater ground locomotion.
The unique challenges of underwater operations, such as reduced gravity, high water damping, and limited access to full-state measurements of the robot and its surrounding environment, required innovative solutions. Taking inspiration from the literature on the Spring Loaded Inverted Pendulum Model (SLIP), we derived mathematical models in the form of reduced-order dynamical systems capable of describing the dynamics of a reference biological system. These models allowed to study the behavior of the system while spanning different design choices (such as the amount of compliance in the leg), actuation strategies, and environment. Additionally, these methodologies can inform on the system stability and power efficiency. These analyses lead to design choices and to the definition of controllers that could wrap the dynamics of the robotic system to the modeled one.
The study also explores the relationship between terrestrial and underwater SLIP models, emphasizing the role of the medium (air or water) in shaping motor schemes. Notably, reinforcement learning demonstrates the beneficial impact of underwater experience on learning ground locomotion. Overall, this thesis contributes valuable insights and practical applications to advance legged robotics in challenging environments.
Using an existing underwater hexapod robot (SILVER2), we performed extensive tests in the field of the developed controllers integrated with various behaviors, showcasing underwater operations like sediment sampling and litter collection.
Despite the remarkable progress in the field, several challenges limit the widespread use of legged robotics in field operations, including locomotion control and stability, power efficiency, and terrain adaptability.
The work carried out in this thesis focused on addressing these limitations through mathematical modeling grounded on biomechanical studies. Significant emphasis was placed on adapting the capabilities of terrestrial legged systems to novel environments, such as underwater ground locomotion.
The unique challenges of underwater operations, such as reduced gravity, high water damping, and limited access to full-state measurements of the robot and its surrounding environment, required innovative solutions. Taking inspiration from the literature on the Spring Loaded Inverted Pendulum Model (SLIP), we derived mathematical models in the form of reduced-order dynamical systems capable of describing the dynamics of a reference biological system. These models allowed to study the behavior of the system while spanning different design choices (such as the amount of compliance in the leg), actuation strategies, and environment. Additionally, these methodologies can inform on the system stability and power efficiency. These analyses lead to design choices and to the definition of controllers that could wrap the dynamics of the robotic system to the modeled one.
The study also explores the relationship between terrestrial and underwater SLIP models, emphasizing the role of the medium (air or water) in shaping motor schemes. Notably, reinforcement learning demonstrates the beneficial impact of underwater experience on learning ground locomotion. Overall, this thesis contributes valuable insights and practical applications to advance legged robotics in challenging environments.
Using an existing underwater hexapod robot (SILVER2), we performed extensive tests in the field of the developed controllers integrated with various behaviors, showcasing underwater operations like sediment sampling and litter collection.
File
| Nome file | Dimensione |
|---|---|
Ci sono 1 file riservati su richiesta dell'autore. |
|