These physical factors play an important role in the quality of digital twin models, especially the sensor outputs (LiDAR, camera, radar). The common practice of manually assigning those properties in the world model is expensive, time consuming, and not scalable.
We are working to automate this process by combining advanced AI technologies with Simcenter Prescan capabilities of 3D world and sensor modeling. Please check below our recent demo using the nuScenes traffic dataset, where material and texture properties have been integrated into the 3D simulation model. When validated with physics-based sensors, the simulated sensor outputs closely resemble real data.
With great contribution from Andy Huynh, João Malheiro Silva, Hamid Reza Abdolhay, Pieter Sibma, Freddy Mullakkal, Serkan Ergun,… and our academic collaborator: Prof. Holger Caesar (TU Delft).
And some similar activities and demos:
- 3D Munich city: https://lnkd.in/grg-ptic
- Gaussian Splatting: https://lnkd.in/eejD3DRf
- Real2Sim of nuScenes, Waymo dataset: https://lnkd.in/eDhvGPbq