Munal: test the spacecraft and learn from the outcome

Munal was a one-unit CubeSat project involving high-school students with APN and the Nepal Academy of Science and Technology. The team’s 2023 presentation described an in-house electrical design, imaging, a combined computing-and-radio experiment, and an art-and-culture contribution. Its proposed image-selection software illustrates a practical problem: a spacecraft may acquire more observations than it can afford to transmit.

ISRO’s PSLV-C62 brochure lists Munal among the co-passenger payloads. The rocket launched on 12 January 2026, but the mission failed; ISRO reported an anomaly at the end of the third stage, and India’s Department of Space later confirmed the failure review. Munal did not reach its intended orbit. The cited evidence does not identify the students’ satellite as the cause. Preflight test results remain engineering evidence, while planned orbital measurements must not be presented as achieved results.

Consider an invented image-selection test containing one hundred labeled pictures: sixty useful scenes and forty unusable scenes. A classifier keeps fifty-four useful scenes and six unusable ones. Sixty pictures are selected, of which fifty-four are useful, so selection precision — open definition is ninety percent. Six useful scenes were also lost. A single success percentage cannot tell the whole story.

If the selected pictures have equal file sizes, transmitting sixty rather than one hundred reduces this test’s image-data load by forty percent. That saving must be weighed against the missed useful observations. Labels, representative test conditions and a separate evaluation set matter: a rule tuned and judged on the same familiar pictures may look better than it performs on new clouds, lighting or landscapes.

Audit the images that are kept and those that are lost

A team can retain design files, ground test records, software and the reasoning behind component choices after a launch failure. Those materials can support another mission, but only if the configuration is clear. A laboratory image classifier working on recorded scenes has passed a particular ground test; it has not thereby demonstrated reliable operation through radiation, temperature changes and real orbital viewing conditions.

For a follow-on design, write the requirement in measurable terms: how many useful scenes may be missed, what data reduction is needed, and how an operator can detect unexpected behavior. Keeping a small sample of rejected images could help diagnose bias, at the cost of additional downlink data. This is an example design option, not a claim about Munal’s implemented flight software. Learning from a mission includes being precise about what the evidence can and cannot establish.