This study systematically evaluates the mode-I (opening) and mode-II (shearing) interlaminar strength and fracture toughness of four co-cured fiber–metal laminates (FMLs): AL–CF (aluminum–carbon fiber fabric), AL–GF (aluminum–glass fiber fabric), AL–HC (aluminum–carbon/glass hybrid fabric), and AL–HG (aluminum–glass/carbon hybrid fabric). Epoxy adhesive films were interleaved between metal and composite plies to enhance interfacial bonding. Mode-I interlaminar tensile strength (ILTS) and mode-II interlaminar shear strength (ILSS) were measured using curved beam and short beam tests, respectively, while mode-I and mode-II fracture toughness ($G_{Ic}$ and $G_{IIc}$) were obtained from double cantilever beam (DCB) and end-notched flexure (ENF) tests. Across laminates, interlaminar tensile strength (ILTS) values lie in a narrow band of 31.6–31.8 MPa and interlaminar shear strength (ILSS) values in 41.0–41.9 MPa. The mode-I initiation ($G_{Ic,init}$) and propagation ($G_{Ic, prop}$) toughnesses are 0.44–0.56 kJ/m² and 0.54–0.64 kJ/m², respectively, and the mode-II toughness ($G_{IIc}$) is 0.65–0.79 kJ/m². Scanning electron microscopy reveals that interlaminar failure localizes predominantly at the metal–adhesive interface, displaying river-line features under mode-I and hackle patterns under mode-II, whereas the adhesive–composite interface remains intact. Collectively, the results indicate that, under the present processing and test conditions, interlaminar strength and toughness are governed by the metal–adhesive interface rather than the composite reinforcement type, providing a consistent strength–toughness baseline for model calibration and interfacial design.

Simultaneous localization and mapping (SLAM) has always been a research hotspot in the field of mobile robots. It uses the camera as the main sensor to model the environment during the movement, aiming to help the agent perceive the surrounding environment. Traditional SLAM research mainly focuses on static or low-speed moving objects. For the dynamic scene, because there is no prior information about the environment when estimating its motion, the accuracy and robustness of the system cannot meet the requirements of practical applications. In recent years, due to the development of deep learning technology, visual SLAM has made an unprecedented breakthrough. This paper will discuss recent developments in the use of deep learning for visual SLAM and provide an overview of the key findings from studies that replaced one or more standard SLAM and semantic SLAM modules with deep learning. Finally, the future development of deep learning-based visual SLAM is considered.