GOTO: Gravitational-wave Optical Transient Observer
- Gravitational-wave Optical Transient Observer (GOTO) is a wide-field robotic telescope network designed to identify rapid optical counterparts to gravitational-wave events and other transient phenomena.
- It employs a dual-mode strategy by combining rapid alert-response with regular sky surveys from two antipodal sites using modular arrays of small telescopes.
- Advanced scheduling, automated data pipelines, and machine-learning-based candidate vetting enable GOTO to efficiently convert large-area alerts into precisely localized optical transients.
The Gravitational-wave Optical Transient Observer (GOTO) is a dedicated, wide-field, robotic optical telescope network built to identify optical counterparts to gravitational-wave detections and other rapidly evolving transients. In its mature configuration, GOTO operates from two antipodal sites—La Palma in the Canary Islands and Siding Spring in Australia—with two independent robotic mounts at each site; each mount is formed by an array of eight 40 cm telescopes with a field of view of 44 square degrees, giving 88 square degrees per site. The network is intended to survey the visible sky every 2–3 days and to provide near-24-hour response to transient alerts within a minute of their detection (Dyer et al., 2024).
1. Scientific rationale and observational niche
GOTO was founded in 2014 as a collaboration spanning institutions in the United Kingdom, Australia, Thailand, Spain, and Finland, in direct response to the observational mismatch between gravitational-wave localization areas and the speed with which optical counterparts fade (Dyer et al., 2020). That mismatch was already evident in early multi-messenger campaigns: GW170817 was localized to roughly 30 square degrees, whereas GW190425 had an initial skymap exceeding 10,000 square degrees, and later assessments noted that future median localizations were still expected to remain in the thousands of square degrees (Dyer et al., 2022). For kilonovae and related merger-powered transients, the consequence is operational rather than merely instrumental: a facility must cover large areas quickly enough to image the source before it disappears.
A common misconception is that GOTO is only an alert-response instrument. The project’s design is explicitly dual-mode. It combines rapid reaction to transient notices with a regular wide-field sky survey that builds the reference-image archive required for difference imaging and candidate validation (Dyer et al., 2020). This survey-plus-follow-up architecture places GOTO at the boundary between discovery facilities and large-aperture follow-up observatories: it is intended to transform a coarse gravitational-wave, gamma-ray, or neutrino localization into a precisely localized optical transient suitable for spectroscopy and dense multi-wavelength follow-up (Dyer et al., 2024).
Although the project was conceived primarily for gravitational-wave counterpart searches, the operational alert stream is broader. Facility descriptions include LVK gravitational-wave notices, Fermi-GBM, Swift-BAT, GECAM, and IceCube alerts, and later scientific papers show GOTO functioning as a discovery engine for gamma-ray-burst afterglows as well as for serendipitous transients in survey mode (Dyer et al., 2024).
2. Network architecture and hardware evolution
GOTO’s central engineering idea is modularity: instead of a single large telescope, each robotic mount carries an aligned array of eight 40 cm unit telescopes, allowing the field of view to scale approximately linearly with the number of units and mounts. Early papers described the unit telescopes as fast Wynne-Newtonian astrographs with an focal ratio, On-Semi KAF-50100 CCDs, pixels, and $1.25$ arcsec/pixel sampling; later system descriptions specify 40 cm H400 Wynne–Riccardi astrographs on direct-drive DDM500 German equatorial mounts, with Baader filters and FLI ML50100 cameras containing 50 megapixel KAF-50100 CCDs (Dyer et al., 2020, Dyer et al., 2024). In the 2024 network description, each unit telescope covers , and the eight telescopes on a mount are tiled with slight overlap to form a combined 44 square-degree footprint (Dyer et al., 2024).
The observatory infrastructure was designed to minimize overheads. Each mount is housed in an AstroHaven clamshell dome, a configuration chosen to avoid dome-rotation waiting time and to support rapid slewing between tiles during transient response (Dyer et al., 2020).
| Epoch | Configuration | Reported field of view |
|---|---|---|
| Jul 2017 | 1 mount, 4 UTs | 18 sq deg |
| Aug 2020 | 1 mount, 8 UTs | 40 sq deg |
| 2024 | 2 mounts per site, 2 sites | 44 sq deg per mount; 88 sq deg per site |
The prototype saw first light in June 2017 and was inaugurated on 3 July 2017 at the Observatorio del Roque de los Muchachos on La Palma (Dyer, 2020, Dyer et al., 2020). The initial GOTO-4 configuration was later upgraded to a full 8-unit-telescope array in 2020, with a second northern mount funded in 2019 and a southern node planned at Siding Spring Observatory (Dyer et al., 2020). By 2024 the network was described as fully operational at both antipodal sites (Dyer et al., 2024).
3. Robotic control, safety systems, and scheduling
The enabling software layer is G-TeCS, the GOTO Telescope Control System, a distributed Python-based robotic control stack that separates hardware access from higher-level autonomy. Its core components are hardware daemons, a master “pilot” process, a conditions daemon, a sentinel alert listener, an observation database, and a just-in-time scheduler (Dyer et al., 2020). The hardware daemons cover cameras, exposure queue, filter wheels, focusers, OTA hardware, dome, mount, and power control, while the pilot orchestrates the night through asynchronous coroutines that manage startup, calibration frames, flats, autofocus, science observations, and shutdown (Dyer et al., 2020).
Safety logic is built into both software and hardware paths. The conditions daemon monitors rain, temperature, humidity, dew point, wind speed, wind gusts, internal temperature and humidity, network status, free disk space, UPS state, and dome hatch status, converting these inputs into simple state flags. If the sum of flags is greater than zero, the dome closes and the pilot pauses (Dyer et al., 2020). The dome daemon is unusual among the daemons in that it can act autonomously, specifically to force closure during unsafe conditions (Dyer et al., 2020).
Target selection is intentionally dynamic. Rather than computing a fixed nightly plan, the pilot queries the scheduler at high cadence and receives the currently best pointing under visibility constraints and priority rules. Prototype descriptions state that the scheduler was queried every 10 seconds; a later facility overview described reprioritization every 5 seconds, but both accounts characterize the scheme as “just-in-time” scheduling (Dyer et al., 2020, Dyer et al., 2022). A key ranking term is the effective rank
where is the starting rank and is the number of successful previous observations. For equal effective rank, the scheduler uses a tie-break that combines tile weight and airmass ,
0
This structure allows repeated visits to be interleaved across multiple active events without letting a single tile monopolize the facility (Dyer et al., 2020).
The alert-ingestion side is similarly specialized. Sentinel listens to transient notices, classifies them through GOTO-alert, and inserts the resulting pointings into the observation database. Recognized packet types include Swift BAT and Fermi GBM GRB alerts, along with LVC preliminary, initial, update, and retraction notices (Dyer et al., 2020). The broader operational picture is that each telescope can act independently under central coordination, so failures on one mount do not halt the rest of the network (Dyer et al., 2022).
4. Survey strategy and measured follow-up performance
GOTO’s sky survey and follow-up modes are tightly coupled. The survey populates a fixed tiling grid with reference images, while alert-mode observations revisit the same geometrical tiles so that subtraction against recent templates can be carried out immediately (Dyer et al., 2020). Survey descriptions evolved with the instrument. Prototype-era papers emphasized a standard sequence of 1 s exposures in the broad 2 filter spanning 400–700 nm, with typical limiting magnitudes of 19–21 for the three-exposure stack and a representative depth of 20th magnitude in three minutes (Dyer et al., 2020). The 2024 network overview described the standard survey as four 45 s exposures in 3, reaching about 20 mag in dark time (Dyer et al., 2024).
Prototype survey operations already demonstrated the cadence logic behind the reference archive. One assessment reported that the survey observed each of approximately 2200 visible grid tiles, with each tile observed on average 20 times over 18 months (Dyer et al., 2022). This reference library is operationally critical because GOTO’s transient-discovery model depends on difference imaging rather than on catalog-only novelty tests.
The O3a gravitational-wave campaign with the GOTO-4 prototype provides the clearest quantified performance benchmark. During the first half of O3, GOTO followed 29 GW superevents. No viable electromagnetic counterpart candidate was identified, but the system delivered a fastest response of less than 1 minute when the source region was observable immediately, an average first-observation delay of 4 hours after alert receipt, and a mean tiled area of 5 square degrees per event. This corresponded to 6 of the LVC probability map on average, or 7 of the observable probability, with the largest single area reaching 8 square degrees (Gompertz et al., 2020). The same analysis concluded that even the prototype could detect AT2017gfo-like kilonovae beyond 200 Mpc in favourable observing conditions (Gompertz et al., 2020).
Later full-network operations indicate the expected reduction in response overheads. The 2024 facility paper stated that, under normal conditions when the target is visible, GOTO can begin observations within 30 seconds of alert receipt; for the candidate event S240422ed, GOTO started observations within 3 minutes of the trigger and covered 9 of the skymap within the first night (Dyer et al., 2024). This suggests that the observatory’s scientific utility scales not only with field of view but also with the maturity of its scheduling, transfer, and vetting infrastructure.
5. Data reduction, transient vetting, and machine-learning infrastructure
GOTO’s discovery capability depends on a low-latency end-to-end pipeline rather than on telescope hardware alone. The 2026 workflow paper describes an ETL-style architecture in which raw FITS images are transferred from the sites to Warwick, orchestrated by Apache Airflow with Celery workers and PostgreSQL-backed metadata, processed into calibrated images and difference-image products, and exposed through the Marshall for automated and human vetting (Lyman et al., 2 Mar 2026). The pipeline typically completes about 7 minutes after shutter close. Raw transfer from La Palma had $1.25$0 s and $1.25$1 s; from Siding Spring the corresponding values were $1.25$2 s and $1.25$3 s (Lyman et al., 2 Mar 2026).
Image processing includes standard CCD reduction, astrometric and photometric calibration, set-image generation by stacking single exposures, and difference image analysis against deep historical templates. The subtraction stage uses HOTPANTS, while candidate validation combines forced photometry, multiplicity across constituent singles, and a CNN-based real-bogus classifier (Lyman et al., 2 Mar 2026). Candidates crossing the Marshall thresholds—real-bogus $1.25$4 in survey mode and $1.25$5 in event-follow-up mode—enter a triage system of Pending, Inbox, Stream, Store, Junk, and Banish. Over a representative period, automated vetting moved about 8158 candidates per day to Junk, about 1610 per day to Store, and about 123 per day to Inbox, leaving collaboration members to inspect roughly 100 objects per day (Lyman et al., 2 Mar 2026). GOAT, the GOTO Auto-Trigger system, can submit follow-up requests about 11 minutes after shutter close (Lyman et al., 2 Mar 2026).
Rubin Observatory LSST Science Pipelines were independently adapted to GOTO data as a secondary processing route. Using the obs_goto package, the coaddition pipeline achieved sub-pixel astrometry, with a mean angular separation of 0.31 arcsec relative to Pan-STARRS DR1, photometric accuracy of about 50 mmag at $1.25$6 and about 200 mmag at $1.25$7, and a coadded $1.25$8-band depth of roughly 19.6 mag (Mullaney et al., 2020). A related forced-photometry study found precision typically better than 20 mmag for sources brighter than 16 mag, agreement with colour-corrected Pan-STARRS ranging from 10 mmag for bright sources to 200 mmag for faint sources, and a $1.25$9 0-band survey depth between 19 and 20 magnitudes depending on observing conditions (Makrygianni et al., 2021).
On the classification side, GOTO has also served as a testbed for sequential ML on sparse light curves. A mixed-input recurrent neural-network classifier using photometric time-series plus contextual inputs such as Galactic coordinates and distance to the nearest GLADE galaxy achieved an AUC of 0.972 with weighted focal loss on a three-class problem comprising variable stars, AGN, and supernovae (Burhanudin et al., 2021). Since July 2023, the Kilonova Seekers citizen-science project has been integrated into this broader vetting ecosystem; the facility overview states that it has produced multiple discoveries of real sources that would otherwise have been missed and has generated training data for new automated classifiers (Dyer et al., 2024). A plausible implication is that GOTO’s operational identity is now inseparable from its human-in-the-loop and ML-assisted selection stack.
6. Scientific results, limitations, and role in multi-messenger astronomy
In gravitational-wave astronomy, GOTO’s significance lies less in a single detection than in persistent infrastructure-building around rapid wide-field optical search. The prototype O3a campaign established that a 4-unit-telescope system could tile hundreds of square degrees per alert and detect AT2017gfo-like kilonovae beyond 200 Mpc in favourable conditions, even though no viable counterpart was identified in that sample (Gompertz et al., 2020). Later facility papers frame the full dual-hemisphere array as sufficiently wide-field and autonomous to cover large localization regions while also supplying the survey references needed for robust subtraction (Dyer et al., 2024).
The project’s broader scientific record already extends well beyond gravitational-wave triggers. By May 2024, GOTO was reporting more than 100 transients per month to the Transient Name Server, and the 2026 pipeline paper reported 4623 first discoveries by GOTO on TNS between 2019-05-08 and 2026-01-28, with 11800 total reported transient candidates (Dyer et al., 2024, Lyman et al., 2 Mar 2026). These figures show that the observatory is functioning as a general transient-discovery facility rather than solely as a dormant GW-response instrument awaiting a rare counterpart.
Gamma-ray-burst afterglows provide the clearest published examples of this broader role. For GRB 250818B, GOTO began observing 0.54 hours after the trigger, obtained 1 s exposures in its wide 2 band, and identified a fading optical source absent in pre-trigger GOTO imaging taken 9.23 hours earlier down to 3 mag. The afterglow was measured at 4 mag at 0.54 h and had faded to 5 mag by 1.67 h; that localization enabled subsequent spectroscopy with Keck/LRIS and a broadband interpretation at 6 (Belkin et al., 18 Feb 2026). A separate 2025 study presented seven long-GRB afterglows discovered by GOTO in 2024, with six spectroscopic redshifts spanning 7 to 8, reinforcing the facility’s effectiveness for poorly localized high-energy transients (Kumar et al., 11 Sep 2025).
Another misconception is that wide-field robotic facilities only provide crude first detections. In GOTO’s case, the scientific chain runs from rapid identification through contextual vetting and onward to coordinated follow-up on facilities such as the Liverpool Telescope, pt5m, and GOTO-FAST on the Isaac Newton Telescope (Dyer et al., 2024). The observatory’s role is therefore intermediary but technically indispensable: it converts low-latency, large-area alerts into observationally tractable astrophysical sources. Within modern multi-messenger astronomy, that function is not ancillary; it is the operational bottleneck that makes detailed downstream science possible.