Hi everyone,

I'd like to share a project I've been developing for the past couple of years: HYBRID Telescope Control v1.2.0 Pro, a Windows desktop application (WPF / .NET 4.8 / C# 8) that tries to bring together everything a serious imager or visual observer needs — hardware control, imaging pipeline, plate solving, adaptive tracking, AI assistance, and voice commands — under a single roof.

The centerpiece of the design is a full ASCOM layer: the app talks to your mount, camera, focuser, filter wheel, rotator, dome, weather station, safety monitor, and more through standard ASCOM COM drivers or ASCOM Alpaca (network), so it works with essentially any hardware your existing drivers already support.

What ASCOM device classes does it cover?

Nine full ASCOM device classes are implemented:

Device class	Notes
Telescope (mount)	GoTo, tracking, pulse-guide, park/unpark, sync
Camera	Mono, CFA, DSLR, also UVC/webcam via ASCOM
Focuser	Absolute + relative, temperature compensation
Filter wheel	Auto filter sequencing per target
Rotator	Hardware derotation loop for Alt-Az
Dome	Slaved to mount Az, auto-shutter
ObservingConditions	Weather feed to EKF + refraction model
SafetyMonitor	Hard abort + park on unsafe flag
Switch	Flat panel, dew heaters, power distribution

The app also speaks LX200 serial for legacy mounts and exposes its own ASCOM Alpaca server so other tools (KStars, SGP, Voyager…) can connect to a virtual mount or camera it controls.

Tracking & pointing: the Alt-Az EKF story

This is where I spent the most R&D time. Alt-Az mounts are inherently harder to track than equatorial ones: the azimuth rate blows up near the zenith (proportional to 1/cos(Alt)), field rotation is constant, and standard guiding software wasn't designed with these non-linearities in mind.

HYBRID addresses this with a 6-state Extended Kalman Filter:

• State vector: [Az, Az_dot, Alt, Alt_dot, bias_Az, bias_Alt]
• Process model: non-linear sidereal equations + numerical Jacobian
• Measurement updates: plate-solve corrections at 1–30 s intervals; pulse-guide corrections at 1–2 Hz
• Zenith Guard: inflates Q_az above 85°, hard clamp at 89.5° — prevents the singularity from destabilizing the filter
• Innovation gating: χ² test at 99% confidence (2 DOF) — bad plate solves are rejected before they corrupt the state
• ML-Enhanced EKF: a small neural net trained on session drift history fuses with the EKF (70/30 weighting) for predictive correction
• Field derotation loop: the rotator is commanded in real time by the EKF's field-rotation integral

On top of the EKF, a classic TPoint-style 6-term pointing model (IH, CH, NP, MA, ME, FO) is fitted from plate-solve data and applied at slew time. This gives the triple-correction stack: pointing model → EKF tracking → PEC FFT.

Voice control + AI (Claude)

80+ voice commands are recognized through an embedded STT pipeline. Examples by category:

GoTo & targeting

• "Go to Andromeda Galaxy", "Slew to NGC 7000", "Go to M13"
• "Go to coordinates RA 5h 34m, Dec +22°"

Imaging

• "Start sequence", "Pause capture", "Take 10 frames 120 seconds", "Change filter to Ha"

Focusing

• "Auto focus", "Move focuser in 50 steps", "Temperature compensation on"

Tracking & guiding

• "Start tracking", "Calibrate guiding", "Dither now"

Observatory

• "Open dome", "Park telescope", "Check weather", "What's the dew risk?"

AI assistant (Claude via MCP)

• "What's the best target for tonight?", "Explain this histogram", "Analyse this frame"

The AI layer runs through 5 MCP servers, meaning Claude (or any MCP-compatible LLM) can invoke telescope functions directly — slew, capture, plate-solve, check status — as native tool calls. This is what makes the voice+AI loop feel natural: you can have a conversation about an object while the telescope is already moving to it.

Imaging pipeline

The full pipeline from photons to finished image is built in:

• Calibration: Bias, Dark, Dark-Flat, Flat; Auto-Dark-Optimization (ADO) to reuse darks across temperatures; per-Bayer-channel flat normalization
• Alignment: SEP star detection → triangle pattern matching → ASTAP / Astrometry.net plate-solve fallback
• Normalization: Median + MAD location/scale matching; local per-tile background normalization (Siril-style)
• Rejection (9 algorithms): Iterative σ-clip, Winsorized σ-clip, Percentile, Linear Fit Clip, GESD, and more
• Combination: Mean / Median / Sum / Multi-Frame Deconvolution / Drizzle (mono + CFA) / Mosaic stitching
• Color: SPCC using Gaia DR3 photometry; narrowband combiner (SHO, HOO); SCNR
• Stretching & detail: Auto-stretch MTF / Asinh / GHT; RBF/polynomial background extraction; à trous wavelets; Anscombe VST denoising; Richardson-Lucy deconvolution

Other highlights

• Scheduler: Full multi-target session planning with altitude/Moon-separation constraints, automatic meridian flip (EQ) or zenith avoidance (AltAz), priority ranking
• Sky chart: Embedded live sky chart with real-time target overlays
• Optical diagnostics: Star FWHM, eccentricity, Strehl ratio estimation, coma/astigmatism pattern analysis
• Remote observatory: Cloud sync, remote session handover, INDI/Alpaca bridge for Linux/Mac clients
• Licence: GPL-3.0 core; Pro commercial license for the full feature set (free during the current launch period at hybrid-telescop-control.com)

Stack

Layer	Technology
UI	WPF / .NET 4.8 / C# 8
Image processing	Python 3.11.9 (NumPy, Astropy, photutils, scikit-image)
ASCOM	COM Automation + Alpaca REST
Plate solve	ASTAP, Astrometry.net
AI / voice	Claude via MCP; embedded STT
Platform	Windows 10/11 x64

Comparison snapshot

Compared to the usual suspects:

• vs. NINA / SGP / Sequence: HYBRID adds the EKF tracking layer and full AI/voice control; SGP/NINA don't have native Alt-Az EKF compensation.
• vs. PHD2: PHD2 is a dedicated guider (excellent for sub-pixel OAG guiding). HYBRID EKF can replace PHD2 for Alt-Az mounts when plate-solve rate is ≥ 1/min, but PHD2 remains the better choice if you have an OAG and want sub-pixel guiding.
• vs. MountWizzard4: MW4 shines with 10micron HW7/AZ firmware. On generic ASCOM mounts (EQMOD, OnStep, iOptron), HYBRID EKF works without any hardware dependency.
• vs. TPoint: TPoint corrects where you aim (static pointing model). HYBRID's EKF corrects how you track (dynamic). HYBRID includes both.
• Release comming up soon,
• Happy to answer questions about the ASCOM implementation, the EKF math, or the MCP/AI integration. Clear skies!