Why Robotics Is Hard: The 'Fall Damage' of Sim2Real

generativist · x · 2026-08-05

Amid recent discussions in the AI community about the immense difficulty of robotics, the author offers an evolutionary perspective: the primary reason biological evolution took so long is that sim2real is real, and reality operates at a fixed clock rate.

If we attempt to accelerate this process and substitute it with gradient descent, we inevitably take "fall damage" when bridging the sim2real gap, due to the irreducible complexities of the physical world.

Original post →

More from Embodied

Embodied channel →