All Tracks are now accepting applications for the new cohort. Apply via the Get Involved page. Collaboration enquiries are open year-round.
Space AI Research Group

Building trustworthy
AI for the
final frontier

We develop autonomous, interpretable AI systems for space missions — from onboard inference on resource-constrained CubeSats to formal verification of mission-critical decision-making. Pre-deployment validation, not post-incident response.

View Research Get involved →
0
Members
0
Countries
0
Publications
IAC 2025 Oral Presentation
Preselected · Acta Astronautica

Why we exist

Pre-Deployment Validation

Our methodology centers on interpretable AI validated before launch. We engineer systems that explain their decision pathways — not after something goes wrong, but before it ever could.

Full-Stack AI Engineering

From ML model architecture through deployment pipelines — covering onboard inference, satellite health monitoring, and Earth observation analytics across hardware constraints and radiation-hardened computing.

Mission-Critical Safety

Formal verification for AI in life-critical systems. Mars rover decision-making, satellite collision avoidance — mathematical proofs of behavior bounds, not just security hardening.

Explainability as Requirement

XAI in space isn't academic — when a satellite autonomously changes orbit, operators need to understand why. We're building the language for human-AI communication in space operations.

Quantum-Classical Hybrid Computing

Our QML division explores quantum algorithms for remote sensing and optimization in orbital mechanics — at the intersection of theoretical computer science and practical space engineering.

Hands-On Mission Integration

Real hardware, real launches, real data. Debugging edge computing on 10cm³ satellites, optimizing power budgets for neural networks, handling intermittent ground station contacts.

Not Hype. Not Flashy.

Reliability. Explainability. Accountability. We are not building AI that impresses — we are building AI that you can trust when the margin for error is zero.

AI That Helps, Not Replaces

We don't build AI to replace humans in space exploration. We build AI to help them operate in the most vulnerable environment humanity has ever attempted — where there is no margin, no fallback, and no second chance.

SATISH Architecture

Spacecraft Autonomous Telemetry Intelligence and Safety Handling — a modular AI safety system for CubeSats that detects failures, flags anomalies, recommends corrective actions, and explains every decision.

MISSION CONTROL

Ground Station

Mission Ops

Human Operator

Manual Override & Approval
Satellite Subsystems

COMM

Propulsion

ADCS

EPS

Sensors

Flight Computer

OBC

Telemetry Module

Data Acquisition
Onboard Data
Risk Prediction Layer

Anomaly Detection

LSTM / Autoencoder
Anomaly Flags

Failure Prediction

XGBoost / TST
Decision Module

Decision Support System

PPO RL Agent

Ethics & Policy

UNOOSA Guideline Guardrails

Resource Reconfiguration Engine

Execution & Command
Accuracy: 89.5–94.1%
Platforms: 1U · 3U · 6U CubeSat
XAI Models: SHAP + LIME Integration
Governance: UNOOSA · COSPAR Compliant