Ulas Gunes

3D scene reconstruction and digital twinning are revolutionizing industrial processes, enabling smarter, more efficient operations. My research focuses on Gaussian splatting, a cutting-edge method for reconstructing complex 3D environments from visual data. Unlike traditional methods, Gaussian splatting excels at creating highly accurate, real-time 3D models of industrial spaces and equipment with minimal computational overhead.

This technology opens up exciting opportunities for the industrial sector, particularly in areas like electrical machine diagnostics, automated manufacturing, and factory supervision. By creating precise digital twins—virtual replicas of physical systems—manufacturers can monitor and analyze operations in real-time, predict equipment failures, and optimize processes without disrupting production. In agriculture, 3D reconstructions can map fields and monitor crops, supporting precision farming and sustainable practices (John Deere is using this technology in their tractors for example).

In automation, these high-fidelity 3D reconstructions can enhance robotics by enabling more precise navigation and interaction in dynamic factory environments. For electrical machines, the technology can support advanced condition monitoring and maintenance planning by visualizing wear, alignment, or system inefficiencies. Beyond industrial applications, this technology extends to consumer sectors, such as enhancing virtual reality experiences in platforms like Meta’s VR headsets, where accurate 3D environments improve immersion and usability. The ability to quickly reconstruct scenes also supports applications like remote training, virtual factory tours, and collaborative design.

My research not only makes 3D scene reconstruction faster and more accessible but also bridges the gap between cutting-edge computer vision and practical industrial applications. The need for expensive and time consuming machine learning techniques (and hardware such as cameras and sensors) for training autonomous vehicles is reduced significantly, thanks to this technology (Volvo is using this technology to train their cars in simulation environments, rather than the real world for example, so that the model training can continue 24/7, and different driving conditions can be created).