Vision-based Cost Maps for Safe Autonomous Navigation : Design and Evaluation of Vision-based Control Barrier Functions
Raja, Golnaz (2023)
2023
Master's Programme in Computing Sciences
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2023-11-13
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202311029353
https://urn.fi/URN:NBN:fi:tuni-202311029353
Tiivistelmä
Designing safety-critical systems for unfamiliar environments is a substantial challenge in the field of robotics. Control Barrier Functions (CBF) serve as a common tool for addressing this challenge. However, the definition of CBF based on perceptual input remains a relatively unexplored and complex area of research.
This thesis extends prior work, where the authors, including the author of this thesis, introduced the innovative concept of Vision-based Control Barrier Functions (V-CBF). V-CBF defines Control Barrier Functions using perceptual input obtained from an RGBD camera, enabling the avoidance of obstacles with arbitrary shapes in unknown environments. A pivotal element of V-CBF is the 2D customized cost map, which transforms the segmented unsafe sets into an appropriate format that satisfies the requirements of CBF. The design and evaluation of these cost maps play a crucial role in the proper generation of V-CBFs. However, this aspect has not been thoroughly extensively in previous research.
The thesis embarks on a comprehensive investigation of diverse methodologies for generating these essential cost maps. The proposed methods are rigorously implemented and assessed within the CARLA simulator. To offer a thorough evaluation, both qualitative and quantitative comparisons are conducted, drawing from industry-standard ISO 22737 guidelines and custom-designed metrics within the CARLA simulator environment.
Furthermore, to substantiate the practical applicability of V-CBF, it is implemented on an industrial mobile robot. This real-world deployment serves as a tangible demonstration of the effectiveness of V-CBF in unknown environments, emphasizing its potential beyond simulated contexts. The transition from simulation to tangible real-world implementation underscores the portability and robustness of V-CBF, signifying its relevance and promise in real-world scenarios.
This thesis extends prior work, where the authors, including the author of this thesis, introduced the innovative concept of Vision-based Control Barrier Functions (V-CBF). V-CBF defines Control Barrier Functions using perceptual input obtained from an RGBD camera, enabling the avoidance of obstacles with arbitrary shapes in unknown environments. A pivotal element of V-CBF is the 2D customized cost map, which transforms the segmented unsafe sets into an appropriate format that satisfies the requirements of CBF. The design and evaluation of these cost maps play a crucial role in the proper generation of V-CBFs. However, this aspect has not been thoroughly extensively in previous research.
The thesis embarks on a comprehensive investigation of diverse methodologies for generating these essential cost maps. The proposed methods are rigorously implemented and assessed within the CARLA simulator. To offer a thorough evaluation, both qualitative and quantitative comparisons are conducted, drawing from industry-standard ISO 22737 guidelines and custom-designed metrics within the CARLA simulator environment.
Furthermore, to substantiate the practical applicability of V-CBF, it is implemented on an industrial mobile robot. This real-world deployment serves as a tangible demonstration of the effectiveness of V-CBF in unknown environments, emphasizing its potential beyond simulated contexts. The transition from simulation to tangible real-world implementation underscores the portability and robustness of V-CBF, signifying its relevance and promise in real-world scenarios.