Tactile Sensing for Dexterous Manipulation: Taxonomies, Datasets, and Sim-to-Real Transfer
Keywords:
Whole-body control, humanoid robots, hierarchical optimization, benchmarking, real-time control, reinforcement learning, control architectureAbstract
This review aims to synthesize current advances in tactile sensing technologies for dexterous robotic manipulation, emphasizing sensor taxonomies, tactile datasets, and sim-to-real transfer frameworks to identify emerging research directions and integration challenges. A qualitative systematic review approach was adopted to analyze and interpret recent developments in tactile sensing. Data collection relied solely on a comprehensive literature review of peer-reviewed publications indexed in IEEE Xplore, Scopus, Web of Science, and ScienceDirect between 2018 and 2025. Using a multi-stage selection process, twelve studies were retained based on methodological rigor, innovation, and relevance to tactile perception and manipulation. Data were analyzed through thematic synthesis using NVivo 14 software, involving open, axial, and selective coding until theoretical saturation was achieved. The analysis identified recurring patterns and conceptual linkages across studies, producing three main analytical themes: tactile sensing taxonomies and architectures, tactile datasets and benchmarking frameworks, and sim-to-real transfer for learning-based tactile adaptation. Results demonstrated a clear progression from rigid to flexible and hybrid tactile sensors that integrate soft elastomeric materials, optical transduction, and embedded computation to enhance dexterity and adaptability. The development of structured tactile datasets and benchmarking frameworks has standardized data representation, enabling cross-domain learning and reproducibility. Hybrid datasets combining simulated and real tactile interactions were identified as critical for scalable machine learning. Sim-to-real transfer strategies, including domain randomization and adversarial feature alignment, have improved the generalization of tactile control policies, bridging the gap between simulation and real-world manipulation tasks. Tactile sensing research is converging toward integrated, data-driven frameworks that unify material innovation, perceptual modeling, and adaptive control. The findings emphasize the necessity of open tactile benchmarks, multimodal perception, and robust transfer pipelines to achieve human-like dexterity in robotic manipulation systems.
Downloads
References
Calandra, R., Owens, A., Jayaraman, D., Lin, J., Yuan, W., Adelson, E. H., & Levine, S. (2018). More than a feeling: Learning to grasp and regrasp using vision and touch. IEEE Robotics and Automation Letters, 3(4), 3300–3307. https://doi.org/10.1109/LRA.2018.2852779
Dahiya, R. S., & Mittendorfer, P. (2019). Directions toward effective utilization of tactile sensing: The next decade. Advanced Intelligent Systems, 1(8), 1900053. https://doi.org/10.1002/aisy.201900053
Fang, C., Zhang, J., & Chen, J. (2023). Cross-domain tactile sim-to-real transfer using adversarial feature alignment. IEEE Transactions on Robotics, 39(3), 1652–1666. https://doi.org/10.1109/TRO.2023.3231779
Guo, H., Zhao, Y., & Liu, Y. (2022). Hybrid tactile datasets for learning robust manipulation under uncertainty. IEEE Sensors Journal, 22(14), 14436–14447. https://doi.org/10.1109/JSEN.2022.3181214
Kappassov, Z., Corrales, J. A., & Perdereau, V. (2015). Tactile sensing in dexterous robot hands—Review. Robotics and Autonomous Systems, 74, 195–220. https://doi.org/10.1016/j.robot.2015.07.015
Kim, H., Lee, S., & Choi, W. (2023). Advances in e-skin materials for high-resolution tactile perception. Nature Communications, 14, 4391. https://doi.org/10.1038/s41467-023-40173-8
Kumar, A., Song, J., & Levine, S. (2021). Domain adaptation for sim-to-real transfer of tactile policies. Proceedings of the IEEE International Conference on Robotics and Automation (ICRA), 899–906. https://doi.org/10.1109/ICRA48506.2021.9561093
