Six minutes and fifty-four seconds after liftoff, a 71-metre rocket stage came back down from 96 km, lit 13 engines less than 2 km above the ground, slid sideways into the gap between two steel arms and set its catch pins down on them. These eight sheets take the catch apart, from the decision to fly without legs to the last few metres.
Falcon 9 lands on four legs that fold out in the last seconds of flight. Super Heavy has none. At the end of 2020 Elon Musk said SpaceX would try catching the booster with the launch tower instead, and the idea went through three design revisions before Flight 5 proved it.
“We’re going to try to catch the Super Heavy Booster with the launch tower arm, using the grid fins to take the load.”
“Saves mass & cost of legs & enables immediate repositioning of booster on to launch mount — ready to refly in under an hour.”
Elon Musk on X · 30 Dec 2020 SRC
The trade is mass for infrastructure. Legs ride the booster to staging and back on every flight, and the booster must carry propellant to lift them and to land them. The tower’s arms never leave the ground. The cost moves to the pad: it has to work every time, and the booster has to arrive at a target the size of a pair of steel rails.
| Rev | Date | Change | Source |
|---|---|---|---|
| A | — | Landing legs on the booster (the Falcon 9 approach) | baseline |
| B | 30 Dec 2020 | Delete legs. Catch on the tower arm, grid fins take the load | Musk on X |
| C | Aug 2021 | Two pins for lifting and catching; the fins stop carrying the load | Musk, via Inverse |
| D | 13 Oct 2024 | First catch, Booster 12 on Flight 5 | SpaceX |
The tower is a 146-metre steel lattice. A carriage rides tracks on three of its four legs, lifted by cables that run over a sheave at the top and down to a winch at the base. The carriage carries two arms, roughly 36 m long, that swing open and shut on hydraulic actuators. The same arms stack the rocket before launch and catch the booster after it.
| Item | Part | What it does | Source |
|---|---|---|---|
| 1 | Tower | 146 m (479 ft) steel lattice, the tallest launch tower on record | Guinness SRC |
| 2 | Carriage | Rides tracks fixed to three of the four tower legs on 12 guide blocks | Teslarati 2021 SRC |
| 3 | Drawworks | Winch at the base; cable passes over the crown sheave and through a travelling block (5 wire-rope passes on Pad A) | NSF 2025 SRC |
| 4 | Chopstick arms | About 36 m long by an observer’s measurement; swing open and shut | Teslarati 2021 EST |
| 5 | Arm actuators | Linear hydraulic actuators; shielded before Flight 5 | Wikipedia, NSF SRC |
| 6 | Landing rails | The pins land on them; they compress as shock absorbers | NSF 2024 SRC |
| 7 | Bumpers | Compressible steel on the arms’ inner faces, replacing foam and rubber; the booster slides down them | NSF 2024 SRC |
| 8 | Catch pins (×2) | Hardpoints near the top of the booster, between the grid fins | Wikipedia, SpaceNews SRC |
| 9 | Jackscrews + stabilizer pins | After the catch, translate the booster to line it up for lowering | NSF 2024 SRC |
| 10 | Launch mount | The arms lower the booster back onto it after the catch | NSF 2024 SRC |
The webcast showed two booster numbers on every frame: altitude in whole kilometres and speed in km/h. Everything on this sheet starts from those readings, OCR’d at 30 frames per second. The path through the sky is reconstructed from them, and it lands back at the tower within about 0.1 km after an 88 km excursion, which is a good sign the method holds. Drag across the charts or press play.
SpaceX wrote the rule down before launch. A catch would be tried only if thousands of vehicle and pad criteria were met, automated checks found the booster and the tower healthy, and the Flight Director sent a manual command before the boostback burn ended at T+3:41.
“If this command is not sent prior to the completion of the boostback burn, or if automated health checks show unacceptable conditions with Super Heavy or the tower, the booster will default to a trajectory that takes it to a landing burn and soft splashdown in the Gulf of Mexico.”
SpaceX, Flight 5 mission page SRC
The default is the safe one. Doing nothing sends the booster to the water, and the tower has to be actively chosen. The checks stay armed after the command: software would abort the catch if either the booster or the chopsticks reported something unhealthy.
“Lost comms to the launch tower computer. Catch would probably still have worked, but we weren’t sure, so erred on the side of caution.”
Elon Musk on X · Nov 2024 SRC
At staging the booster was about 67 km up, still climbing, and moving downrange at roughly 1.24 km/s. It flipped, relit the ten-engine middle ring alongside the center three, and in under a minute turned that into about 0.5 km/s heading back toward Texas. The thrust dropped sharply ten seconds before shutdown, which looks like a trim on fewer engines.
Then the engines went quiet for nearly three minutes. Super Heavy flies no entry burn. It coasted over a 96 km apogee, fell back to 4,395 km/h, and let the air brake it. Below 20 km the deceleration reached about 5.7 g with no engines running, and the dynamic pressure peaked at several times what the full stack saw on the way up.
The booster came in on a slant. Its speed on the webcast is higher than its rate of descent, so some of that speed was sideways. At landing-burn ignition the reconstruction puts it roughly 0.8 km east of the tower, 1.8 km up, moving west at about 150 m/s. The 13-engine phase killed the sideways motion in about seven seconds; the last 500 m were close to vertical on the center three.
NASASpaceflight described the same move from the video: the booster “translated over to get in between the chopsticks.” A likely reason to come in from the side is that the booster isn’t pointed at the tower until its engines are lit and steering. Falcon 9 uses the same idea on return-to-launch-site landings: it aims just off the coast, so a failure ends in the water.
The last 20 metres took about seven seconds. In that time the arms closed, the booster slid down the bumpers until its pins met the landing rails, and the rails compressed as the engines shut off. The HUD speed went from 37 km/h to zero, and at the very end it was falling at walking pace.
Frame times and speeds are HUD readings. When exactly the arms began closing isn’t published; frames 2–4 are placed by the NASASpaceflight description of the sequence EST.
A simplified model of the last 1.8 km, started from the reconstructed Flight 5 state at landing-burn ignition. Let the autopilot fly it and throw faults at it, or take the controls yourself at the 13-to-3 engine switch. Then run a few hundred attempts and see where they land. The tolerances are my estimates; SpaceX hasn’t published them.
| Quantity | Where it comes from | Trust |
|---|---|---|
| Speed | Webcast HUD, OCR’d frame by frame (SofieBrink/StarshipTelemetryExtractor). 1 km/h resolution. | HUD high |
| Altitude | Same source, but only whole kilometres. The smooth curve is fitted through the moments the digit changes, which pins it to about ±0.1 km; at the end, height comes from integrating speed. | HUD + CALC |
| Downrange path | Horizontal speed = √(speed² − climb rate²), integrated. Closes on the tower within 0.12 km out of 88 km; I removed that residual as a linear drift. | CALC good shape, ±0.3 km late |
| Mach, q, g-loads | Derived from the path with a standard atmosphere. Loads near apogee are noisy because altitude is coarse up there. | CALC medium |
| Propellant | No published numbers. Rocket equation run backwards from an assumed catch mass through the burns measured in the telemetry. The band spans dry mass 200–275 t and 5–35 t left at the catch. | EST low |
| Event times | SpaceX’s planned timeline, Wikipedia’s as-flown times, NSF reporting, and what the telemetry shows. Sources disagree by 1–3 s. | SRC ±3 s |
| Engine counts | 33 → 3 at MECO, 13 for boostback, 13 → 3 in the landing burn (NSF, SpaceX). The trim at T+3:31 and the 13→3 switch time at T+6:38 are read from the deceleration. | SRC + CALC |
| Tower and booster dims | Tower 146 m (Guinness); booster 71 × 9 m (SpaceX); arm ≈36 m (observer measurement). Arm spacing, rail length, pin size and catch height are drawn schematically. | SRC / EST |
| Sandbox tolerances | Mine. Chosen to be physically sensible, not published by SpaceX. | EST |