DWARFLAB opens Draco pre-orders around a 90 mm automated imaging stack
What changed
DWARFLAB's Draco product page is now taking pre-orders for a 90 mm integrated smart telescope. The manufacturer lists a 1/1.3-inch, 50 MP telephoto sensor that uses 1.2 micrometre native pixels for solar-system work and 2-by-2 binning for roughly 12 MP deep-sky images with 2.4 micrometre effective pixels. It advertises guided single exposures up to 300 seconds in an integrated equatorial mode, automatic focus, calibration, tracking and live stacking, FITS and TIFF export, scheduled remote operation, and automated mosaics. Its built-in H-alpha plus O III filter is specified at 13 plus or minus 3 nm per band; the SHO edition adds S II plus O III. These are pre-shipping manufacturer claims. The official page gives equivalent focal length rather than the physical focal length and active sensor dimensions needed for a dependable field-of-view preset.
Why it matters
This design collapses telescope, camera, guider, filter system, calibration library, mount and processing pipeline into one product. A database record therefore cannot treat it like a conventional optical tube paired with an interchangeable camera. Sampling changes by mode, filters change by edition, and a quoted maximum sub-exposure says little about repeatable yield under wind, seeing, polar-alignment error or sky brightness. Built-in flats, bias data and real-time darks also make calibration provenance and export behaviour part of the equipment specification.
Southern-sky relevance
The equatorial workflow must be verified against the south celestial pole and South African latitudes, and its target library should be checked on southern emission nebulae and large mosaics. Solar safety differs by edition: the Standard model lists a switchable internal ND filter, while the SHO edition requires an external filter to be fitted manually. That distinction is critical before any unattended solar schedule is suggested.
AstroClassifieds tool opportunity
Add a smart-telescope data model spanning optical mode, native and binned pixel pitch, effective resolution, physical focal length, active sensor size, filter bandpasses, mount mode, advertised maximum sub-exposure, calibration method, raw export and solar-filter type. Prepare a Draco candidate record for the telescope and camera databases, field-of-view tool and exposure planner, but keep it labelled pre-release and do not enable calculated framing until physical focal length and sensor dimensions are confirmed. Store advertised and independently measured performance separately.
Worth reading
NASA and IBM release an open lunar foundation model and task benchmarks
What changed
On 10 September, NASA announced the NASA-IBM Lunar Foundation Model, trained primarily on Lunar Reconnaissance Orbiter data. NASA says the training set contains roughly two million tiles, including more than one million one-metre camera images and nearly 964,000 multispectral images at 100-metre resolution. Public downstream tasks cover crater detection, irregular mare-patch segmentation and polar-ice prospectivity. Model weights, machine-learning-ready data and benchmark collections are available, and the GitHub repository is Apache-2.0 licensed. Scope caveat: the repository describes the release as inference and fine-tuning code and explicitly says pretraining code is not included.
Why it matters
This is a reproducible starting point for machine-assisted lunar feature analysis rather than a generic image enhancer. It could accelerate crater and terrain annotation, but the training domain is georeferenced orbital imagery with modalities and resolution unlike a seeing-limited amateur video stack. Performance on LRO benchmarks does not establish accuracy on backyard images, and the ice-prospectivity output does not make lunar polar ice directly visible from Earth.
Southern-sky relevance
The Moon is accessible from South Africa, but its apparent rotation, terminator geometry and any mirror reversal from a diagonal or capture train differ from the north-up orbital products used in many maps. A useful southern implementation must transform labels into the observer's actual orientation and account for libration and illumination; otherwise even correct feature detections will be confusing.
AstroClassifieds tool opportunity
Prototype an offline lunar annotation experiment for the planetary processor and sky map: limb-fit or plate-solve a stacked lunar image, transform it to a lunar reference frame, then compare candidate crater labels with an established LRO catalogue. Test first on a small, versioned South African image set spanning phases, seeing and common mirror orientations. Report confidence and provenance, and do not expose model output as authoritative until false positives and orientation errors are measured.
Worth reading
- NASA, IBM Launch AI Foundation Model for Lunar Science NASA Science
- NASA-IBM Lunar Foundation Model code repository NASA IMPACT
- NASA-IBM Lunar-FM and downstream models NASA-IBM AI4Science