I get that it's hard, but not why it's fundamentally harder than beating a grandmaster at chess.

Is the gravel loose or packed? Was there a wet spot that will cause a loss of traction later on when the water on the tire freezes? Is that rut solid or soft? That spot looks frozen but there's standing water around it. Can it eyeball how deep the mud may be in a spot to decide whether it's ok to cross or attempt to maneuver around? There's moose tracks here, be cautious. There is more to it than just "hey let's automate this"

Edit, more: tree down across the road, truck can't drive over it. Get out and wrap a chain around it to move it. Busted axle, how will it fix that? Flat tire? Mechanic used distilled water like a dumbass and it froze. Road sign was taken out by an accident, guardrail is missing, hope you arent using those to determine how to keep on track.

Yup, and you haven't even dealt with one other driver on the road yet.

That's a laughable statement. Chess is a perfect fit for computer problem solving.

Chess involves an opponent that's doing their best to cause trouble.

Driving on icy gravel, that's just a physics simulation, with no opponents.

The hard part of self-driving is vision, with hard constraints on real-time and power supply — how long that takes (especially given the enormous pile of scenarios we don't think about when trying to list them but do know how to deal with when we encounter them), I don't want to guess, as I think it could be anywhere from negative 6 months to positive 16 years.