Interfaces and couplings

Sixty-six interfaces exist between the twelve subsystems, and every pair has at least one. Thirty-seven of the 66 are high criticality, 25 medium and 4 low. That saturation is itself the finding: at this level of integration there is no isolated subsystem change, and the order in which a team freezes its numbers matters more than the order in which it designs its parts.

Twelve nodes, five kinds of interface

Each subsystem is a node that trades in a small set of numbers. Every interface is one of five types, because the five fail in different ways and are owned by different engineers.

Timing is an interface type rather than a management afterthought because the competition scores it. The astrobiology clock is 20–30 min, the autonomy-to-servicing transition can be as short as 10 min, and setup is 15 min including the weigh-in and all communications S-URC27-RULES S-URC27-QA §1.a.ii–§1.e.i EQ4. A design that is two minutes too slow loses the same points as a design that does not work.

The 12 × 12 matrix

Each cell is the interface as seen from the subsystem in that row. The letters give the interface type, in descending order of dominance, followed by its criticality: H high, M medium, L low. The upper triangle is the authoritative entry and the lower triangle repeats it so the table can be read in either direction. The diagonal is a subsystem’s own business and is marked n/a.

Five readings follow from the matrix. The two most coupled subsystems are the arm and perception at nine high interfaces each, one mechanical and one data-dominated. Seven subsystems sit at six or seven, so there is no clean partition into core and periphery. Delivery has a single high interface, which is evidence that its difficulty is in the mission sequence rather than in a novel coupling. The astronaut node is data and timing dominated, with the arm as its only high mechanical interface. Every power interface carries an electrical term, and the only one without it is the thermal link to the environment node, because the E-stop rule makes one subsystem carry a system function S-URC27-RULES §2.a.v.

Eight interfaces with a number that forces a decision

The full register is 41 rows across the five types: ten mechanical and stow, seven electrical, ten data, six thermal and eight timing. Every high cell in the matrix is represented by at least one row. The table below searches and filters by interface type.

The three chains that carry the most consequence

Chain 1: arm mass to drivetrain to points. The mass loop is stable and converges in two iterations with a loop gain of 0.0137 kg of battery per kilogram of rover, so forgetting to re-size the pack after adding the arm is wrong by 0.16 kg. The danger is two cliffs. At 250 mm wheels, a 45° loose slope and 45 kg requires 9.97 N·m per wheel, and an AK80-9 is rated 9.0 N·m S-CUBEMARS-AK80-9. Inverting the formula gives 40.6 kg. The 14 kg ARES-class arm lands at 40.3 kg, 0.3 kg below the cliff, so one more bracket forces a 0.94 kg actuator class at +0.45 kg per wheel or +2.7 kg for six, re-opening the budget at 43.0 kg with 7.0 kg of headroom. The arm is not 4.5 kg lighter than the alternative; it is 4.5 kg plus the actuator class it preserves. No published team process owns the margin-to-cliff number.

Chain 2: wheel diameter is a four-way, irreversible coupling. It changes climbing, centre-of-gravity height, stow length and torque in conflicting directions. From 250 to 510 mm costs 12.9° of static tip angle, against 5.6° for an arm change from 4.4 to 14 kg. Six printed wheels grow from 6.0 to 25.0 kg, torque at 40 kg on a 30° slope grows from 6.9 to 14.0 N·m, and overall length grows from 1.15 to 1.41 m. Only top speed improves, from 5.1 to 10.4 m/s at rated motor speed, against a design target of 0.5–2.0 m/s.

The field evidence agrees. WVU ran 20-inch wheels in 2024, 24-inch in 2025 and back to 20-inch with no folding mechanism in 2026, giving up 2 inches of climbing radius to fit the cube without a fold S-WVU-SAR2025 S-WVU-SAR2026. Missouri S&T moved to a suspension that no longer needs to fold to fit the 1.2 m box S-MST-SITE-ATHENA. Two of the strongest teams independently concluded that geometry beats mechanisms. The defensible point for a first build is 250–300 mm with a 0.85–0.95 m wheelbase and no fold, which leaves 40–50 mm per side for suspension links and tolerance.

Chain 3: camera count is a link-budget decision that converts into stage time. Six cameras at 720p30 and 3 Mbit/s each demand 18 Mbit/s. The 900 MHz link carries 200 kbps at 1 km, which is 1.1 % of the demand; 2.4 GHz carries 6 Mbit/s at 2.62 km and 54 Mbit/s at 0.42 km; 5.8 GHz with the radio at the mast carries 6 Mbit/s at 21.4 km. Dropping from the fastest modulation to 6 Mbit/s buys about 9 times the range. At 1 km and 0.5 m/s the Delivery traverse takes 33 minutes, which does not fit a 30-minute stage, and WVU’s 2026 tri-band system was tested to only 750 m line of sight and 164 m obstructed S-WVU-SAR2026. The camera count must be frozen against the link: three or four streams, adaptive bitrate, and a separate stills path.

Two shorter chains. Arm reach against stow and typing is not a horizontal problem: a 1.05 m arm folds to 0.55 m and fits with 0.65 m to spare, but the parked vertical budget is only 0.12–0.15 m, and the fold joint that solves it adds backlash at the worst lever. At 4.4–5.2 mrad, the shoulder’s 1.05 m lever gives 4.6–5.5 mm of lost motion at the tip, four joints sum to 9.7–11.4 mm, and the typing target is ±3 mm against a 5.0 mm geometric limit. The resolution is a wrist camera and visual servoing, which turns the shoulder term into 1.8 mm at a 0.35 m lever, not a low-backlash reducer at every joint.

Mast height against stability and the cube is the other. The stowed stack already uses 0.84 m of the 1.2 m envelope before the mast, so a 2.0 m mast leaves 0.06 m and a 2.5 m mast fails. The same 30 mm tube has its first mode at 19 Hz at 0.8 m, 4.8 Hz at 2.0 m and 2.6 Hz at 3.0 m, inside the 2–15 Hz terrain band. A 20 mm tube is at 9 Hz at 0.8 m. The resolution is two masts: a tall guyed radio mast and a short stiff camera post, the only fold a first team should keep.

Conflicts that no subsystem can settle

Fourteen decisions cannot be made inside one subsystem, because each has an owner on both sides. The table gives the two sides and the review by which the answer is needed.

The freeze order follows the penalty, not the subsystem number

The interfaces whose consequence is a rule penalty are frozen first; the rest are a re-test. A worse but frozen choice beats a better but late one, which is also the judges’ own advice on design stability S-URC-ADVICE.

Open problems

• No team publishes a mass, power or centre-of-gravity budget, so every mass-distribution number in the chains is an engineering estimate with its stated basis.
• The weigh-in duration is not published, which is one reason not to build a mission plan around a fast re-weigh.
• Teleoperation ground speed and glass-to-glass latency are not measured by any team; both are inputs to the camera-chain arithmetic and both need a field test.
• Whether a drone counts toward the 70 kg total, and whether a spare drone battery falls under the one E-stop rule, is not stated in the rules and changes a design S-URC27-RULES §2.b.vii.
• Whether the Equipment Servicing cache is the team’s own astrobiology cache is open, and it changes both the cache and the end-effector design.

Further reading

• Mass and power carries the mass and pack numbers these chains consume.
• How rovers fail shows which of these interfaces actually breaks on the field.
• Coupling and trades works the fourteen conflicts as trade studies.
• Systems engineering owns the allocation and the review gates.