Robotic systems are designed to extend our reach into space and other extremely harsh environments, to reconnaissance unexplored worlds or areas that are not accessible by humans, and to manipulate assets and resources. The operations of these systems are strongly limited by a low degree of automation, the onboard computational capacity and the need of human interaction and intervention. Advances in sensing and perception are fundamental to enhance the performance of assets devoted to human and robotic exploration.
Methods and tools for visual data processing are of critical importance in the design of robotic systems that are hosted onboard spacecraft and surface, aerial, and underwater vehicles. The measurements acquired by the sensors are analyzed to retrieve the desired information regarding the asset operation to better determine its pose and to enhance the path planning for future tasks. A conventional technique to perform a planned task consists in the combination of data and theoretical/semi-empirical models. The localization of planetary rovers, for example, is obtained through algorithms that process range finder data or stereo-images by accounting for an accurate modeling of the kinematic and/or dynamical equations. The navigation of spacecraft is carried out by analyzing deep space radio tracking data that are acquired by ground stations. To enhance the trajectory reconstruction of orbiters, additional data including altimetric and optical measurements are processed and combined with radio observations.
These methods, however, are computationally expensive, preventing from an autonomous control loop because of the limited onboard CPU resources. Another important weakness of these conventional algorithms is the lack of knowledge of the physical laws that rule a certain task. The application of AI-based algorithm for the navigation and path planning of deep space probes and assets devoted to the exploration of harsh environments is then fundamental to enable the implementation and development of reliable and safe autonomous systems.
Artificial intelligence (AI) addresses the main issues related to the Guidance, Navigation and Control (GNC) through alternative algorithm schemes that are based on neural networks (NN). Supervised and unsupervised Machine Learning (ML) algorithms adopt a preliminary NN training with and without pairs of labeled input/output, respectively. Unsupervised learning yields the detection of natural patterns in the dataset processed in the NN training. These techniques are well-suited for space and robotic applications whose dynamical model is poorly known.
Reinforcement learning is an interactive ML paradigm that learns a control policy end eventually learns a system transition function (i.e., a dynamics model), to guide its definition. A preliminary knowledge of the dynamical model that predicts the evolution of the true robotic system is not always necessary, but can be helpful to generate low-cost control policies, enabling accurate convergence of the algorithm. Decision making and operations of robotic assets are also cast as multi-class classification problems. Deep Learning (DL), and, in particular, Convolutional Neural Networks (CNNs) have proven to be the most effective architecture to deal with image classification and feature extraction.
The main scope of this Research Topic is focused on the application of AI techniques for the Guidance Navigation and Control (GNC) system of spacecraft and surface, aerial and underwater vehicles. Methods for autonomous navigation of orbital and surface assets are encouraged. Potential paper topics include, but are not limited to:
- Use of Computer Vision (CV) for prototypes of the navigation systems of spacecraft and surface, aerial and underwater vehicles;
- Deep learning-based algorithms for terrain classification and obstacle avoidance;
- ML algorithms for feature detection on extremely harsh environments;
- VSLAM algorithms for path planning;
- Model-Based Reinforcement Learning (MBRL) for autonomous navigation.
Topic Editor Roberto Capobianco is employed by Sony AI. All other Topic Editors declare no competing interests with regards to the Research Topic subject.
Keywords:
Deep Learning, Planetary Rover, Navigation, Path planning, Computer Vision, Guidance Navigation and Control (GNC) System
Important Note:
All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.
Robotic systems are designed to extend our reach into space and other extremely harsh environments, to reconnaissance unexplored worlds or areas that are not accessible by humans, and to manipulate assets and resources. The operations of these systems are strongly limited by a low degree of automation, the onboard computational capacity and the need of human interaction and intervention. Advances in sensing and perception are fundamental to enhance the performance of assets devoted to human and robotic exploration.
Methods and tools for visual data processing are of critical importance in the design of robotic systems that are hosted onboard spacecraft and surface, aerial, and underwater vehicles. The measurements acquired by the sensors are analyzed to retrieve the desired information regarding the asset operation to better determine its pose and to enhance the path planning for future tasks. A conventional technique to perform a planned task consists in the combination of data and theoretical/semi-empirical models. The localization of planetary rovers, for example, is obtained through algorithms that process range finder data or stereo-images by accounting for an accurate modeling of the kinematic and/or dynamical equations. The navigation of spacecraft is carried out by analyzing deep space radio tracking data that are acquired by ground stations. To enhance the trajectory reconstruction of orbiters, additional data including altimetric and optical measurements are processed and combined with radio observations.
These methods, however, are computationally expensive, preventing from an autonomous control loop because of the limited onboard CPU resources. Another important weakness of these conventional algorithms is the lack of knowledge of the physical laws that rule a certain task. The application of AI-based algorithm for the navigation and path planning of deep space probes and assets devoted to the exploration of harsh environments is then fundamental to enable the implementation and development of reliable and safe autonomous systems.
Artificial intelligence (AI) addresses the main issues related to the Guidance, Navigation and Control (GNC) through alternative algorithm schemes that are based on neural networks (NN). Supervised and unsupervised Machine Learning (ML) algorithms adopt a preliminary NN training with and without pairs of labeled input/output, respectively. Unsupervised learning yields the detection of natural patterns in the dataset processed in the NN training. These techniques are well-suited for space and robotic applications whose dynamical model is poorly known.
Reinforcement learning is an interactive ML paradigm that learns a control policy end eventually learns a system transition function (i.e., a dynamics model), to guide its definition. A preliminary knowledge of the dynamical model that predicts the evolution of the true robotic system is not always necessary, but can be helpful to generate low-cost control policies, enabling accurate convergence of the algorithm. Decision making and operations of robotic assets are also cast as multi-class classification problems. Deep Learning (DL), and, in particular, Convolutional Neural Networks (CNNs) have proven to be the most effective architecture to deal with image classification and feature extraction.
The main scope of this Research Topic is focused on the application of AI techniques for the Guidance Navigation and Control (GNC) system of spacecraft and surface, aerial and underwater vehicles. Methods for autonomous navigation of orbital and surface assets are encouraged. Potential paper topics include, but are not limited to:
- Use of Computer Vision (CV) for prototypes of the navigation systems of spacecraft and surface, aerial and underwater vehicles;
- Deep learning-based algorithms for terrain classification and obstacle avoidance;
- ML algorithms for feature detection on extremely harsh environments;
- VSLAM algorithms for path planning;
- Model-Based Reinforcement Learning (MBRL) for autonomous navigation.
Topic Editor Roberto Capobianco is employed by Sony AI. All other Topic Editors declare no competing interests with regards to the Research Topic subject.
Keywords:
Deep Learning, Planetary Rover, Navigation, Path planning, Computer Vision, Guidance Navigation and Control (GNC) System
Important Note:
All contributions to this Research Topic must be within the scope of the section and journal to which they are submitted, as defined in their mission statements. Frontiers reserves the right to guide an out-of-scope manuscript to a more suitable section or journal at any stage of peer review.