Ex-Tesla Engineer: Pure Vision Policy Training Is the Real Sim2real Challenge

Ex-Tesla engineer Yacine argued that RNNs can learn system identification and that training robot policies directly from vision is not that hard, while a commenter countered that sim2real is easy with sysid but the real difficulty lies in vision-based policy training.

2026-09-03 ~ 2026-09-03 · 2 related posts