Space: guidance, navigation and attitude control with scarce data

Guidance and control from orbit through re-entry and recovery

What happens. A capsule has to leave orbit, survive the heat and the deceleration, and land where recovery can reach it. Each phase has its own model and its own limits, the atmosphere is never the one in the tables, and each flight yields few data for the next.

What it costs. A landing far from the target, loads above the design, or a flight that must be repeated. A small team spends months on the guidance for one mission and has little time left to analyse the flight data.

Why the usual fix fails. A trajectory from a simulation for the nominal day does not hold for a real atmosphere. Widening the margins costs mass and landing accuracy. A large, general toolbox takes longer to adapt than to write anew.

How a fixed-price remote diagnosis checks it. From your trajectory models, vehicle data and the data of earlier flights or tests I check how well the guidance can adapt in flight, and how flight data can improve the models. The fixed-price remote diagnosis answers where optimisation and estimation could raise landing accuracy and which tests would show it.

Background: Optimal paths and setpoint plans

Launcher navigation must be tuned on flight data and monitored for integrity

What happens. The navigation filter of a launcher combines inertial sensors, satellite signals and sometimes other sources. After the first flights the filter is tuned on the recorded data. The team has to show that a faulty sensor or a lost signal is detected before the guidance acts on a wrong position.

What it costs. A flight that is lost or aborted through a navigation error, and a long analysis after each flight to understand why the estimate was off.

Why the usual fix fails. Tuning the noise values by hand to match one flight overfits that flight. Strong monitoring limits create aborts without cause. A purchased navigation unit is a black box that gives no insight for the next vehicle.

How a fixed-price remote diagnosis checks it. From recorded flight and test data I check how the estimate behaves, which tuning is supported by the data, and how an integrity check can separate faults from normal noise. The fixed-price remote diagnosis answers what a tuned filter and a documented monitor would need.

Background: Computing unmeasured states for control

Small teams must build the whole flight software and test it in the loop

What happens. A young vehicle company has to produce navigation, guidance, control and the tests that prove them, from simulation to hardware in the loop to a flight. The team is small and the experts on guidance and estimation are few. Everything is needed at once.

What it costs. A schedule that depends on two or three people, tests that are postponed and a first flight with parts nobody could check as thoroughly as intended.

Why the usual fix fails. Hiring is slow. Buying a ready-made package fits one vehicle class. Doing without simulation saves time until the first hardware test shows an error that a model would have found.

How a fixed-price remote diagnosis checks it. From your architecture, models and test plan I check which part of the chain is most at risk and where a model-based design and a simulation check can replace a test on hardware. The fixed-price remote diagnosis answers what could be delivered separately, with documentation your team can continue.

Reaction wheels need control and fault detection

What happens. Satellites turn with reaction wheels that spin up and down. Friction changes with temperature and age, and the wheel speed passes ranges where vibration is strong. A wheel that is wearing out shows it first in small changes of torque and current, which are easy to overlook.

What it costs. Pointing errors, a wheel that fails earlier than planned and, in the worst case, a lost mission or a safe mode that interrupts service.

Why the usual fix fails. A fixed limit on the motor current detects the failure late. A larger wheel with margin adds mass. A second wheel as backup is built for failure but does not tell when to switch.

How a fixed-price remote diagnosis checks it. From wheel test data and telemetry I check how a model of friction and torque can estimate wear and flag faults early, and how the wheel control can stay accurate near the critical speeds. The fixed-price remote diagnosis answers what this would add to your satellites.

Station keeping, momentum and remaining propellant are planned by rules

What happens. A satellite operator keeps spacecraft in position with thruster firings, unloads momentum from the wheels and estimates how much propellant is left. The remaining life is often estimated from a gauge that is not precise, and the manoeuvres follow rules that were made for the average case.

What it costs. Propellant used on unnecessary manoeuvres shortens the life of a satellite that earns money every day. A wrong estimate of the remaining propellant leads to an early retirement or to a late one that risks a failure.

Why the usual fix fails. Safety margins on the propellant gauge cost revenue years. Manoeuvre rules by calendar ignore the actual drift. A new gauge needs a new satellite.

How a fixed-price remote diagnosis checks it. From telemetry (thruster firings, pressures, temperatures, orbit data) I check how well the propellant can be computed and how a plan of manoeuvres could use less. The fixed-price remote diagnosis answers whether the estimate and the optimisation are precise enough to extend the planned life.

Background: Economic predictive control

Lightweight estimation with event cameras is hard to get small and robust

What happens. An event camera reports only the pixels that change, very fast and with little data. It suits a small spacecraft or a drone with little power. The data are unlike a normal image: sparse, noisy, with no picture to look at. Estimating motion and position from them needs methods that few teams know.

What it costs. A navigation function that does not fit on the available processor, or one that works in the lab and fails in poor light or fast motion.

Why the usual fix fails. Converting the events into images and using a standard method throws away the advantage of speed. A heavy learned method needs a processor that the platform does not have. Tuning by trial runs hides the cases where it fails.

How a fixed-price remote diagnosis checks it. From your recorded event data and the target platform I check which estimator fits the processor, how robust it is in the hard cases and what it needs in a test. The fixed-price remote diagnosis answers whether a lean estimator can reach the accuracy you need.

Background: Computing unmeasured states for control

Collision risk between satellites is predicted too coarsely

What happens. Operators receive warnings that two objects may come close. The position data of each object are uncertain, and the uncertainty grows with time. Most warnings are false alarms, a few are real, and the operator must decide days ahead whether to spend propellant on an avoidance manoeuvre.

What it costs. Propellant and operator time for manoeuvres that were not necessary, and the remaining risk of a collision after a manoeuvre that was not made.

Why the usual fix fails. A low warning threshold produces many manoeuvres. A high one accepts a risk. Waiting for better data comes too late. Buying more observations is expensive and not always available.

How a fixed-price remote diagnosis checks it. From the warning data of your fleet I check how the uncertainty is described, whether the estimate of the risk is realistic and how a better estimate changes the number of manoeuvres. The fixed-price remote diagnosis answers where a better prediction pays for itself.

Background: Computing unmeasured states for control

Pointing performance and sensor calibration limit what a satellite delivers

What happens. A satellite must point its camera or antenna at a target with a small error. The error comes from the sensors for attitude, from the alignment on the structure and from vibration. Calibration happens on the ground, and in orbit things shift with temperature and time.

What it costs. Images or signals that are less sharp than the payload could deliver and calibration campaigns in orbit that take weeks of the operator.

Why the usual fix fails. A better star tracker is a hardware purchase for the next satellite. Manual calibration cannot follow slow changes. Larger pointing margins in the mission plan cost the use of the payload.

How a fixed-price remote diagnosis checks it. From in-orbit data of attitude sensors and payload I check which errors a model can separate from each other and calibrate during the mission. The fixed-price remote diagnosis answers how much pointing accuracy a calibration algorithm can add to the existing hardware.