About this Research Topic
Robots are increasingly being deployed to assist and collaborate with humans in many applications, such as medicine, navigation, and industrial automation. To truly collaborate with humans in complex environments, robots require advanced cognitive capabilities, including the ability to reason with domain-specific commonsense knowledge and the noise observations obtained in the presence of partial observability and non-deterministic action outcomes. Research in Artificial Intelligence (AI) has resulted in sophisticated symbolic formalisms that use temporal and logic relations to represent commonsense domain knowledge, as well as probabilistic, data-driven frameworks that quantitatively represent uncertainty in the decision-making process of robot systems.
Stand-alone symbolic or stochastic AI methods have limitations when applied to robots in complex scenarios. Symbolic AI methods reason with relational descriptions of the attributes of the domain and the robot to guide the robot's behavior. It is, however, often computationally intractable to use these methods to reason about uncertainty quantitatively, or to operate at the level of granularity required for precise interaction with objects in complex domains. Probabilistic and data-driven AI methods, on the other hand, elegantly represent uncertainty quantitatively, and provide mechanisms for reasoning and acting at the level of granularity required for interaction with the physical worlds. These methods, however, offer limited expressiveness for complex cognitive concepts.
There has been considerable interest recently in “hybrid AI” or “neuro-symbolic AI” methods that merge symbolic and probabilistic/data-driven AI methods. For example, recent frameworks have incorporated symbolic domain knowledge to constrain the search space of stochastic planning, and use learned probabilistic models to support more sophisticated symbolic reasoning about concepts such as trust and safety. This research topic seeks to highlight state-of-the-art advances and open problems in this area, in the specific context of developing cognitive systems that enable a robot to accomplish complex tasks.
Contributions to this Research Topic should discuss methods or frameworks for merging symbolic and probabilistic/data-driven AI methods for robotics. Theoretical advances, field reviews, and engineering solutions to real robotics problems are welcome. Potential topics of interest include, but are not limited to:
• Novel ways of integrating symbolic and probabilistic/data-driven representations and reasoning methods (e.g., representation at different levels of abstraction, symbolic reward representation for Markov decision process).
• Improving computational complexity of symbolic-probabilistic reasoning.
• Merging symbolic AI with a probabilistic representation of different sources of uncertainty (e.g., sensor uncertainty, domain description uncertainty, kinematic uncertainty).
• Interesting robotics applications that exploit the benefits of symbolic-probabilistic AI.
Stand-alone symbolic or stochastic AI methods have limitations when applied to robots in complex scenarios. Symbolic AI methods reason with relational descriptions of the attributes of the domain and the robot to guide the robot's behavior. It is, however, often computationally intractable to use these methods to reason about uncertainty quantitatively, or to operate at the level of granularity required for precise interaction with objects in complex domains. Probabilistic and data-driven AI methods, on the other hand, elegantly represent uncertainty quantitatively, and provide mechanisms for reasoning and acting at the level of granularity required for interaction with the physical worlds. These methods, however, offer limited expressiveness for complex cognitive concepts.
There has been considerable interest recently in “hybrid AI” or “neuro-symbolic AI” methods that merge symbolic and probabilistic/data-driven AI methods. For example, recent frameworks have incorporated symbolic domain knowledge to constrain the search space of stochastic planning, and use learned probabilistic models to support more sophisticated symbolic reasoning about concepts such as trust and safety. This research topic seeks to highlight state-of-the-art advances and open problems in this area, in the specific context of developing cognitive systems that enable a robot to accomplish complex tasks.
Contributions to this Research Topic should discuss methods or frameworks for merging symbolic and probabilistic/data-driven AI methods for robotics. Theoretical advances, field reviews, and engineering solutions to real robotics problems are welcome. Potential topics of interest include, but are not limited to:
• Novel ways of integrating symbolic and probabilistic/data-driven representations and reasoning methods (e.g., representation at different levels of abstraction, symbolic reward representation for Markov decision process).
• Improving computational complexity of symbolic-probabilistic reasoning.
• Merging symbolic AI with a probabilistic representation of different sources of uncertainty (e.g., sensor uncertainty, domain description uncertainty, kinematic uncertainty).
• Interesting robotics applications that exploit the benefits of symbolic-probabilistic AI.
Keywords: Symbolic reasoning, Probabilistic reasoning, Reasoning under uncertainty, Hybrid AI, Robotics
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.
