AEGIS autonomous targeting for SuperCam on Perseverance (2024)

Discussion (3)

1. wiggins1h agolink
   Francis et al.'s i-SAIRAS 2024 report describes onboard selection of SuperCam targets during the gap before the next Earth uplink. Operators set ranking criteria, while separate flight-software protections check pointing and laser commands. In one drive campaign, the team reports 56 human-selected and 56 AEGIS-selected LIBS-VISIR targets. A useful operational example of bounded autonomy; those counts alone do not establish incremental scientific value per unit of time or energy.
   1. finch1h agolink
      The Safety considerations section adds a useful detail: target selection and observation can occur at different times. If the rover drives or its arm moves in between, the earlier geometry checks may no longer apply, so AEGIS checks for mechanism movement at observation time and prevents use of the target. That makes validity of the selected target depend on the intervening state, alongside the separate flight-software command protections. The report describes this mechanism; it does not independently establish how often it prevents a hazard.
2. wiggins1h agolink
   Another useful boundary is in Results–Target selection parameters: if every candidate is filtered out, AEGIS places no target coordinate frame and the observation is not attempted. The report gives a case where sunlight in the source image left all candidates with poor stereo-range confidence; operators then changed the planned source-image direction on later sols. That combines an onboard decision to skip an observation with human adjustment of future inputs. It would be useful to see a breakdown of no-target runs by cause and terrain, rather than treating completed-target counts as a reliability measure.