Although 3D reconstruction may seem useful only for graphic and remote visualization of spaces on electronic devices, its application extends beyond aesthetics, especially when it can be performed in real time. In the field of autonomous robotics, volumetric reconstruction plays a fundamental role: by transforming the complexity of physical space into a structured digital model, it allows robots to interpret the environment to carry out critical tasks, such as navigation and path planning. However, generating detailed volumetric maps imposes significant computational challenges, which often exceed the capabilities of standard CPU-based architectures. The need to operate in real time also requires a delicate balance between the quality of the reconstruction and the processing speed.