Neuro-Symbolic Robotics: Capabilities, Limitations, and Challenges
IEEE Transactions on Cognitive and Developmental Systems, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Basım Tarihi: 2026
- Doi Numarası: 10.1109/tcds.2026.3732064
- Dergi Adı: IEEE Transactions on Cognitive and Developmental Systems
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Compendex, INSPEC, Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Anahtar Kelimeler: Cognitive Robotics, Explainable AI, Explainable Robotics, Neural-Symbolic Integration, Neuro-Symbolic AI, Neuro-Symbolic Robotics, Symbol Emergence, Symbol Grounding
- Orta Doğu Teknik Üniversitesi Adresli: Hayır
Özet
This paper examines the emerging field of NeuroSymbolic Robotics by systematically mapping the strengths andlimitations of existing neuro-symbolic robotic systems onto anovel taxonomy. Our taxonomy is motivated by a core questionin cognitive and developmental robotics: how should neuraland symbolic processes be organized to produce robots thatcan learn, develop, and reason as intelligent agents do, giventhat symbols are not passively received but actively constructedthrough embodied sensorimotor experience. With this we aimto identify open challenges and promising research directionstoward robust, fully autonomous, and reasoning-capable robots.The advancements in computational power, robust neural structures, and extensive data have positioned Neural Networks asthe preferred solution for robotic challenges involving emergentbehavior, learning, adaptation, and, more recently, reasoningand communication. Despite these strengths, the deploymentof robots in real-world settings requires properties such asverifiability, explainability, and interpretability, which NeuralNetworks lack. Furthermore, neural network-based models oftenstruggle with generalization and extrapolation, thus restrictingtheir use. Historically, symbolic systems have been integral tointelligent robotics due to their verifiability, explainability, andscalability; however, their manually programmed frameworksoften fail to effectively manage the complexity and diversity ofthe robot’s continuous and high-dimensional environments. Inthis paper, by examining the robotic works that combine, ina variety of patterns, neural networks with symbolic systems,we classify them into four main categories of intertwined (A),coupled (B), neuro-symbolic translation (C), and non-uniformneuro-robotic systems (D), and provide an in-depth analysisto elicit their potential for the future of robotics. Based onthis analysis, we conclude that coupled neuro-symbolic roboticapproaches offer the strongest capabilities for robot control andmotion planning, and hence represent the most capable paradigmfor addressing the breadth of real-world robotic challenges.Neuro-symbolic translation-based robotic studies, by naturallyintegrating safety into learning, provide formal verification andcorrectness guarantees, albeit currently having limited practicalapplicability. Finally, intertwined neuro-symbolic robotics adoptsa cognitively inspired, developmental perspective to bridge neuraland symbolic paradigms; yet, despite its transformative potential,its capacity to trigger an architectural paradigm shift has yet tobe demonstrated.