Sahya AI
Overview
Sahya AI is the proprietary, software-defined flight-control and pilot-assist framework developed by SahyaLabs LLP. Designed as the core intelligence layer for the Sahya CX1 (Technology Demonstrator 1 / TD1) eVTOL air taxi, Sahya AI bridges deep-tech autonomous robotics with stringent aerospace certification standards. Beyond powering SahyaLabs' own aircraft, Sahya AI forms the backbone of an asset-light, high-margin "Aero-SaaS" licensing engine for global Original Equipment Manufacturers (OEMs) seeking safety-certified, human-in-the-loop flight operations.
Key Capabilities & Operational Features
Human-in-the-Loop Co-Pilot Architecture: Sahya AI acts as an Augmented Intelligence safety overlay rather than a complete pilot replacement. The human pilot retains high-level tactical command while Sahya AI manages real-time flight envelope protection, wind shear compensation, and micro-adjustments.
Active Obstacle Shielding (Visual Fencing): Monitors near-field and long-range hazards (such as cranes, power lines, or non-cooperative aircraft). If a path violates safety buffers, Sahya AI creates virtual tactile resistance or auto-executes evasive setpoints.
Dynamic Coaxial Failure Compensation: Built specifically to manage the aerodynamic complexity of coaxial X8 octocopters. If a motor or power bus fails, Sahya AI detects the drop in thrust and instantly recalculates motor-mixing matrices across the remaining active motors to keep the airframe level.
Precision Hover-Lock & Optical Anchoring: Integrates downward optical flow with LiDAR point clouds to "anchor" the craft over vertiport landing pads, counteracting urban thermal updrafts and "toilet-bowl" GPS drift.
Explainable AI (XAI) & Neuroadaptive Feedback: Converts system telemetry into natural language alerts (e.g., checklist verification, course change rationale) to reduce pilot cognitive load during high-stress flight phases.
Federated Learning (FL): This is a core software feature of Sahya AI that enables continuous, decentralized model improvement across active eVTOL fleets without requiring raw sensor data to be continuously offloaded to central servers
