Tesla has used LiDAR to train their vision approach. Their approach is to factor it out of the run-time. "Estimating Object Properties Using Visual Image Data" (US Patent No. 10,930,069, granted February 23, 2021) talks about using auxiliary data to train their models.

There’s an ongoing debate about whether LiDAR is truly essential for building a safe autonomous system. Some argue that it offers capabilities vision-based systems lack, while others see it as just another layer of redundancy—adding more data to help complete the perception pipeline.