Technical Specifications

• Modular design: Our system is designed to be very modular. Depending on the use-cases and needs, we can use only one components or use much smaller models.
• While the actual computation power will depend on the components and features needed, most of our systems and models can run entirely offline on-device (Jetson orin nano or similar), providing real-time response. We can run some limited models on CPU only, while other requires GPUs.
• We are currently looking into creating a package for smaller systems (~10TOPs) (Work in Progress)
• RGB-D cameras can be used to create the 3D scene graph to map and recognize surfaces and objects with enough level of detail for most applications. This might help reduce BoM costs by reducing need for LiDAR.
• When necessary, the scene graph precision can be augmented by using 2D-LiDAR sensors where highly accurate measurements are required or to increase the system robustness against changing lighting conditions.