In the test flight of SN9, it appeared that 1 of 2 engines did not relight upon landing. Why is the transition to vertical not done at a higher altitude where a backup engine could be lit if needed or give longer for a single engine to slow it down?
Notice how it came down quite a bit askew apart from not halting the main rotation of the flip: they need two engines for roll control during the maneuver (you can see the engines gimbaling independently to control roll during SN8's flip). Even if this wasn't so, during the flip they have reduced drag slowing their descent: a slower flip means more downward velocity that Starship would need to brake, and more landing propellant.
About starting the third engine: the engines don't take that long to start, and can shut down very quickly. They likely don't need more altitude to start the third, just a startup process that gets it ready to start, but which shuts it down if the first two work properly. I expect they're not doing that simply because starting two engines is complicated enough, they don't want any unnecessary complexity at this point in development.
Also, starting a third engine isn't necessarily helpful. Of the two landing failures they've had so far, the first was from a fuel tank pressure issue. Trying to start a third engine with a propellant system that's already providing inadequate fuel pressure would only make things worse. We don't know why SN9 failed to light the second engine yet, but it could have been from a similar propellant system issue rather than something wrong with the engine itself. Starting the third engine can only help if the problem was internal to one of the other two engines.
The SpaceX Starship is a very ambitious design still in its very early stages of development. The SN8 and SN9 both had about 7 seconds between completing the flip and landing.
G forces decelerating from around 7 meters/second to 0 in 7 seconds work out to a very survivable 3 Gs.
However, it must be pointed out, sans atmosphere on the moon, and very little on Mars, the late flop/flip may be a theoretically possible but less safe option destined to be removed from the final design.
On Mars, the Starship may enter the atmosphere in the prone position, however (especially being liquid fueled), it may benefit itself by descending stably using a drag device such as a grid fin or parachute, then using rockets to land.
Current Mars landers do exactly that: discarding their heat shields after atmospheric friction sufficiently slows the space craft down. Once the Martian Starship slows down even enough, it would then be able to do its "suborbital thing" by flipping (much higher up), extending grid fins, and guiding to its landing sight. The drag device also insures directional stability as now the rocket is flying "backwards" to land on its tail.
Just as in an airplane, a stabilized approach is much safer, in this case particularly in rate of descent. As the stopping thrust requirement is proportional to the square of Velocity, a 20% difference in vertical velocity requires 44% more vertical distance to reach 0 meters/second.
A tall order for even the fastest computerized system. But this is cutting edge. While suggesting Space X consider a safer three step approach to landing (building on the suborbital Falcon 9 booster technology) as follows:
- Flop (70 meters/second) to intermediate altitude
- Transition to vertical with controlled and stable descent using drag device
- Ignite and check retro rockets (adjust as needed)
- Powered landing (larger landing zone to start)
Update on use of a parachute to control rate of descent (for a BFR)
The Space Shuttle solid fuel boosters weighed around 100 tons empty, comparable to SN9's empty weight of around 130 tons. Their parachutes were deployed at around 360 mph. A parachute system for the SN 10 (and beyond) would only need to match the horizontal "flop" rate of descent in the vertical position, allowing more time for engine restart.
The surface area of the "flopped" SN 10 is (generously) 160 x 30 = 4800 feet$^2$ + fins = (roughly) 6000 feet $^2$.
The Space Shuttle solid fuel main chute was 138 feet in diameter, yielding over 14,000 feet $^2$ of drag area with a much higher drag coefficient!
Imagine that parachute, deployed at 20,000 feet, bringing the craft into a slower vertical descent over a dry lake bed.
If the rockets check out ok, cut the parachute loose for a precision powered landing. If not, cut the rocket loose, and save the passenger capsule with the parachute.
Best of luck for SN 10!
I'm surprised that nobody mentioned yet that another likely cause is that the thrust to weight ratio is too high to allow for an efficient flip "too high up above the ground". Let's take as a basic assumption that Starship requires 2 Raptor engines to land (others have mentioned roll control as one reason why this would be needed). Going off public speculation, 1 Raptor produces at the minimum 880 kN of thrust. Thus, 2 Raptors can make a mass of 175 tons hover. Anything lighter will just go up, not come down, under Earth gravity. But Starship's empty mass is speculated to be around 180 tons. Even if it is higher, something like 200 tons, then the final portion of vertical descent for landing will be relatively slow, and consume probably too much fuel.
SpaceX's strategy has always been to do the "hoverslam" since the Falcon 9 days, so it is not surprising that they want to do the same for Starship. This becomes all the more important when landing on bodies with less gravity like the Moon and Mars, where 2 Raptors will be enough to send far heavier objects up, not down.
Simply put, the two aft wings and two forward canard fins work as control surfaces, to guide the vehicle during descent, only in the horizontal position. The starship is just too large to be guided by grid fins, such as done with the F9.
Bonus trivia: The wings and fins are control by motors which are powered by Tesla Model S and X batteries.