Training Methods and Generation Technologies for Embodied Intelligent Robot Manipulation Skill Models: A Systematic Review
Lianpeng Li1,*, Zhoujun Ruan1, Zhichuang Wang2, Haibo Zhang3, Hang Zhong4, Mingyang Li3, Chunpeng Kang5
1 School of Automation, Beijing Information Science and Technology University, Beijing, China
2 School of Artificial Intelligence, University of Science & Technology Beijing, Beijing, China
3 Beijing Institute of Control Engineering, Beijing, China
4 School of Artificial Intelligence and Robotics, Hunan University, Changsha, China
5 Research and Planning Division, Information Center, Ministry of Agriculture and Rural Affairs, Beijing, China
* Corresponding Author: Lianpeng Li. Email: 
(This article belongs to the Special Issue: Robotics Vision and Thinking)
Computers, Materials & Continua https://doi.org/10.32604/cmc.2026.084092
Received 16 April 2026; Accepted 17 June 2026; Published online 04 August 2026
Abstract
Endowing embodied intelligent robots with dexterous manipulation capabilities is paramount for executing complex, open-ended tasks. These capabilities are foundational to advancing true robotic autonomy, thereby facilitating precision assembly, seamless collaborative operations, and highly specialized maneuvers across diverse industrial and service sectors. Focusing on dynamic, unstructured environments where conventional programmed behaviors prove inadequate, this paper presents a systematic, quantitatively driven review of training methodologies and generation techniques for manipulation skill models within the domain of embodied artificial intelligence (AI). To provide rigorous trend validation, this study conducts a comprehensive bibliometric analysis and quantitative literature evaluation. By empirically mapping research distributions and methodological shifts across recent scholarly works, we substantiate our identification of mainstream research trajectories and developmental paradigms. This data-backed approach ensures the delineated landscape accurately reflects the consensus of the broader academic community, effectively mitigating subjective bias. Leveraging this validated framework, we rigorously analyze the core technical challenges currently hindering widespread deployment. These multifaceted challenges encompass high-dimensional nonlinear motion control, algorithmic sample inefficiencies, suboptimal reward formulations, the inherent semantic-to-physical alignment gap, and the computational constraints dictating large-scale applications. Furthermore, this review comprehensively examines key enabling technologies engineered to circumvent these barriers. Specifically, we systematically detail recent breakthroughs in motion control optimization, the evolution of advanced learning paradigms, complex task planning strategies, and synchronized multi-agent cooperative control frameworks. By critically evaluating these advancements, synthesizing the comparative efficacy of existing methodologies, and dissecting real-world deployment bottlenecks, this review establishes a robust theoretical foundation. Ultimately, it identifies practical, empirically supported pathways for future research to accelerate the transition of dexterous robotic manipulation from constrained laboratories to complex real-world applications.
Keywords
Embodied intelligent robots; motion control; advanced learning methods; complex task planning; mult-agent collaboration
Open Access
Login
Register
Submit a Paper
Propose a Special lssue
Download PDF
Supplementary Material File
Downloads
Citation Tools
