SFT 2026-27 - LLASO Project 4 - AI Predictive Maintenance for Lunar Robots

NASA Reference Name: LLASO-P4-AI-MAINT-2026

Executive Summary

Develop an AI system that reads robot telemetry, predicts failures before they happen, and dispatches at-risk robots to the P5 Repair Garage while rotating a standby unit into service — so the fleet keeps running even when no human technician is immediately available. Deliverable: an AI prototype running on simulated or real robot telemetry, with a dispatch interface that feeds the P5 repair workflow.

Requested By

NASA HUNCH / Kennedy Space Center; NASA AI / Autonomy Programs.

Problem Statement

Lunar robot fleets will operate for 30-60 day crewed missions and then autonomously for months between missions. Without predictive maintenance, undetected failures can ground robots at critical moments when no human technician is immediately available.

Requirements Overview

Collect motor, joint, wheel, battery, and thermal telemetry streams

Detect anomalies and predict failures with configurable thresholds

Generate a work order and route the robot to the P5 Repair Garage

Track a standby fleet and dispatch replacements automatically

Accept a robot self-diagnostic override (the robot can confirm or reject an AI flag)

Operate autonomously during uncrewed mission phases

Major Constraints

Real lunar robot telemetry may not be available — synthetic or scaled data must be used

Communication latency to Earth means the system must make decisions autonomously on-orbit

Multiple robot types (P2 transport, P3 unloading, mining, construction) have different failure modes; some robots are always inside and clean, others are always outside and dirty

Repair Garage (P5) capacity is limited — dispatch must prioritize by criticality

Key Challenges

Generating realistic failure signatures in training data without real hardware

Avoiding false positives that waste P5 garage capacity

Handling simultaneous multi-robot failure events gracefully

Balancing the predictive-maintenance schedule against active mission requirements

Cards

1 — Other Points / Comments

Start with robot telemetry 'logs,' which are more accessible and still highly relevant

Lunatics competition robots at KSC are a potential real-hardware data source (TBD)

Encourage teams to interview a NASA robotics project manager about real failure modes they have encountered

2 — Examples of Excellence

The system correctly predicts a motor failure 15+ minutes before a simulated breakdown using telemetry trend analysis

A live dashboard shows fleet status, the maintenance queue, and predicted time-to-failure per robot

The system manages standby rotation with zero mission gaps during a full simulated mission cycle

3 — Examples of Innovation

Federated learning: each robot contributes to a shared failure model without centralizing raw data

A natural-language interface: the crew asks 'which robots are at risk this week?' and gets a spoken summary

A failure root-cause explainer: the AI not only flags a failure but explains WHY in plain language for the technician

Suggestions for High School Students

Simulated telemetry from an RC robot or a stepper motor is an acceptable data source

The AI model can be a simple threshold / rule system for beginner teams; use machine learning for advanced teams

The dispatch interface can be a dashboard mockup rather than a live system

A single robot type (for example, the transport robot only) is fine for keeping the scope manageable

Please review more info below:

Requirements Spec

File Specification

Other Points / Comments

Examples of Excellence

Samples of Innovation

User Stories