The sense of touch is essential for human's effective physical interaction and provides the fundamental means of touch for machines to perceive and interact with the physical world. The capability to perceive contact with surfaces, manipulate objects, and engage in safe interactions with humans plays a pivotal role. This significance becomes particularly evident in emerging fields such as human-machine interaction, wearable robotics, and bio-inspired soft robotics. In these domains, robotic systems necessitate careful engineering using materials that can either imitate or establish physical interactions with soft human tissue. Great efforts have been dedicated to the advancement of developing tactile sensors, utilizing novel transducing methods and materials. Among all, Optical-based sensors, especially vision-based sensors are attracting increasing research interest due to its advantages of several aspects, e.g., high spatial resolution, immune to electromagnetic noise, and are not suffer from wiring difficulty. However, the practical applications of tactile sensors remain limited. they are still far from acceptable to handle dexterous tasks due to the lack of accurate adequate tactile information (3D force, contact area, pressure, and torsion), reliability, and inefficient intensive data training. The signals obtained from tactile sensors can be noisy, high-dimensional, complex, and contain both relevant and irrelevant information. Consequently, there remains a scarcity of sensor configuration design and signal processing techniques capable of effectively addressing these intricate issues. Enhancements in tactile sensing can be achieved through progress in materials, fabrication techniques, and signal processing.

To overcome the abovementioned limitations, this doctoral thesis introduces a flexible tactile sensor, named ColorTac, drawing inspiration from the color-changing mechanism of chameleon skin. By Leveraging nanometer-scale chameleon skin inspired photonic crystal structures, ColorTac not only achieves unprecedented positioning accuracy (0.015 mm) and force accuracy (0.018 N), dozens of times better than existing sensors, but also provides unique pressure and torsion information upon contact. With the optical design of the tactile skin and sensor configuration, we have modeled and validated the contact-to-color relationship between the mechanical stimuli and the physical information on the raw image. Benefiting from the physically interpretable contact-to-color transduction mechanism, by utilizing simple imaging process techniques, ColorTac directly extract the compression, extension, position, and motion information from color images, by taking which as the preknowledge, it just requires ultrasmall model parameters (~1KB) for training, over three orders of magnitude smaller than conventional methods (at least several MB) while maintaining a better performance.

We showcase ColorTac's versatility in tackling diverse challenging tasks, such as recognizing the state of complexly shaped rolling objects, facilitating real-virtual world interaction, and enabling adaptive high-dynamic robot grasping. To further demonstrate the wide application range of ColorTac sensing system, we design a miniaturized force sensing probe for surgical instrument end effector based on ColorTac sensing mechanisms and provide accurate force feedback during minimally invasive surgery.

In conclusion, this thesis innovates in the concept, mechanism, design, and application of the flexible tactile sensor (ColorTac) to overcome the challenges from both the accuracy, lack of sufficient tactile information, and the low efficiency of data processing. Furthermore, this thesis not only offers a general solution for contact-related operations in the physical world but also has the potential to inspire small dataset-driven machine-learning architectures.