About this Research Topic

Manuscript Submission Deadline 24 November 2022

The recent development of intelligent robots benefits from the fast growth of Artificial Intelligence (AI) -powered approaches, such as fuzzy logic, machine learning, and reinforcement learning. As a variety of learning-based methods are proposed for diverse essential robot benchmarks, there still exists a huge gap between the cutting-edge methods and the desire for reliable execution of sophisticated robotic tasks. This is mainly because the conventional learning framework is highly dependent on large sets of data in which the information that is crucial for model promotion may be quite sparse. As a result, the model training process is usually subject to low efficiency and long duration. Furthermore, once trained, minor adjustments of the model to different patterns are very difficult, leading to the lack of flexibility to the environmental changes. This drawback is especially reflected by the ‘end-to-end’ learning schemes, for which a black-box model is trained for the direct mapping between certain domains.

In between the two extremes of the ‘end-to-end’ learning methods and the completely heuristic approaches, one feasible solution to overcome this drawback is to depict the pattern using a series of simpler models which are organized using a heuristic mathematical structure, instead of a single complicated model. Similar learning schemes have also been referred to as hierarchical learning or curriculum learning. The application of heuristic structures allows the incorporation of expert knowledge in a more efficient manner. The decomposed structure makes the model easy to be adapted to new patterns without learning from scratch, and improves the generalizability of the model to wider domains. These advantages are critical for the reliable and flexible implementation of AI-powered methods for sophisticated robot manipulation tasks.

The main goal of this Research Topic is to collect novel ideas and cutting-edge technologies for using architectural Artificial Intelligence to improve the reliability and flexibility of learning-based methods for sophisticated robot manipulation tasks where the variety and uncertainty of the task patterns are present.

Disregarding the ‘end-to-end’ learning schemes that recognize the entire model as a black box, this call is dedicated to properly incorporating heuristic knowledge by designing a feasible structure that combines several simple models but achieves high generalizability for sophisticated robot tasks.

Related topics of interest include, but are not limited to:

• Modular neural networks: a modularized scheme to construct neural networks for better flexibility and robustness
• Curriculum learning: a series of models that are associated with a certain topology used to improve the efficiency of the learning process
• Data-driven control: a combination of the conventional control methods and machine learning for system identification
• Movement Primitives: a virtual dynamic system or stochastic process that serves as the abstract of a robotic task

The contributions solicited must be related to robotic applications, and in particular to task planning, motion planning, control, navigation, perception, reasoning, fault detection and classification, multi-agent coordination, human modeling, etc. Furthermore, the models used to model the patterns and the structure used to incorporate heuristics must be clarified. Models can be deep neural networks, radius-basis-function neural networks, fuzzy logic, support vector machines, heuristic functions with adaptive parameters, decision trees, dynamic systems, probabilistic distributions, etc. An experimental use case or simulation in a high-fidelity environment should be presented to demonstrate the feasibility and applicability of the method.

We welcome regular papers that contain a complete description of a well-solved problem, short brief papers that propose novel perspectives with simple experimental validation, as well as review papers that summarize the recent advances of architectural Artificial Intelligence and evaluate its potential in the application to robotic engineering.

Keywords: Robot manipulation, architectural learning, machine learning, curriculum learning, heuristic learning, robust learning


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.

The recent development of intelligent robots benefits from the fast growth of Artificial Intelligence (AI) -powered approaches, such as fuzzy logic, machine learning, and reinforcement learning. As a variety of learning-based methods are proposed for diverse essential robot benchmarks, there still exists a huge gap between the cutting-edge methods and the desire for reliable execution of sophisticated robotic tasks. This is mainly because the conventional learning framework is highly dependent on large sets of data in which the information that is crucial for model promotion may be quite sparse. As a result, the model training process is usually subject to low efficiency and long duration. Furthermore, once trained, minor adjustments of the model to different patterns are very difficult, leading to the lack of flexibility to the environmental changes. This drawback is especially reflected by the ‘end-to-end’ learning schemes, for which a black-box model is trained for the direct mapping between certain domains.

In between the two extremes of the ‘end-to-end’ learning methods and the completely heuristic approaches, one feasible solution to overcome this drawback is to depict the pattern using a series of simpler models which are organized using a heuristic mathematical structure, instead of a single complicated model. Similar learning schemes have also been referred to as hierarchical learning or curriculum learning. The application of heuristic structures allows the incorporation of expert knowledge in a more efficient manner. The decomposed structure makes the model easy to be adapted to new patterns without learning from scratch, and improves the generalizability of the model to wider domains. These advantages are critical for the reliable and flexible implementation of AI-powered methods for sophisticated robot manipulation tasks.

The main goal of this Research Topic is to collect novel ideas and cutting-edge technologies for using architectural Artificial Intelligence to improve the reliability and flexibility of learning-based methods for sophisticated robot manipulation tasks where the variety and uncertainty of the task patterns are present.

Disregarding the ‘end-to-end’ learning schemes that recognize the entire model as a black box, this call is dedicated to properly incorporating heuristic knowledge by designing a feasible structure that combines several simple models but achieves high generalizability for sophisticated robot tasks.

Related topics of interest include, but are not limited to:

• Modular neural networks: a modularized scheme to construct neural networks for better flexibility and robustness
• Curriculum learning: a series of models that are associated with a certain topology used to improve the efficiency of the learning process
• Data-driven control: a combination of the conventional control methods and machine learning for system identification
• Movement Primitives: a virtual dynamic system or stochastic process that serves as the abstract of a robotic task

The contributions solicited must be related to robotic applications, and in particular to task planning, motion planning, control, navigation, perception, reasoning, fault detection and classification, multi-agent coordination, human modeling, etc. Furthermore, the models used to model the patterns and the structure used to incorporate heuristics must be clarified. Models can be deep neural networks, radius-basis-function neural networks, fuzzy logic, support vector machines, heuristic functions with adaptive parameters, decision trees, dynamic systems, probabilistic distributions, etc. An experimental use case or simulation in a high-fidelity environment should be presented to demonstrate the feasibility and applicability of the method.

We welcome regular papers that contain a complete description of a well-solved problem, short brief papers that propose novel perspectives with simple experimental validation, as well as review papers that summarize the recent advances of architectural Artificial Intelligence and evaluate its potential in the application to robotic engineering.

Keywords: Robot manipulation, architectural learning, machine learning, curriculum learning, heuristic learning, robust learning


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.

Topic Editors

Loading..

Topic Coordinators

Loading..

articles

Sort by:

Loading..

authors

Loading..

views

total views article views article downloads topic views

}
 
Top countries
Top referring sites
Loading..

Share on

About Frontiers Research Topics

With their unique mixes of varied contributions from Original Research to Review Articles, Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author.