Understand why some decisions must happen onboard

A real-time task has a timing requirement tied to the physical process. A navigation answer arriving after a maneuver decision can be mathematically sound yet operationally useless. The relevant measure is whether the entire sensing, computation and action chain meets its required deadline.

Perseverance used terrain-relative navigation during its 2021 Mars landing. Onboard images were compared with stored terrain information to help determine position and choose a reachable landing target. The short descent sequence and communication delay made waiting for an Earth-based decision unsuitable for those immediate actions.

Consider a fictional 100-millisecond deadline. Acquisition takes 20 ms, processing 50 ms and output delivery 10 ms, totaling 80 ms and leaving 20 ms before the deadline. If competing work adds a 30-ms wait, the total becomes 110 ms and misses the requirement.

This is a teaching budget, not a rover specification. It shows why measuring an isolated algorithm’s average speed is insufficient. Worst-case workload, communication between onboard components and time spent waiting for shared resources can determine whether the deadline is met.

Time the whole chain: fictional budget

A real-time design should specify the response to missing, stale or low-quality input and to a missed deadline. The appropriate fallback depends on the mission phase. Waiting can be reasonable for a parked rover but may be impossible during an irreversible descent sequence.

Evaluate the response in representative scenarios, including interactions with other tasks. A claim that an algorithm is fast enough needs a defined workload and timing evidence. A claim that it is autonomous needs a defined decision scope; neither adjective by itself establishes suitability for the physical operation.