I am thinking of using deep learning to simulate LiDAR sensor data for autonomous vehicles case. After searching all I can find is to use GANs or VAEs but the data generated is static and in the case of AVs the vehicle is moving. The question I ask is can learning based simulation for LiDARs in the case of AVs is possible?
If the images coming from LIDAR are similar (in general say same distribution features) as the ones to be used in validation, testing and in real scenarios then why not not use them!
There have been some mentions recently of generative models that can produce video, which sounds like a similar concept to what you are proposing, since the LiDAR data will change over time like video frames. E.g. here’s a link that Prof Ng gave in one of the recent The Batch articles. Maybe try googling “generative models for video” and see if you find any other examples of work like that. Maybe you can find some ideas that you can adapt for your purposes, but it sounds like pretty deep waters. It’s a really interesting idea. Let us know how it goes! 