Stronger Communities + Scientific Progress

RiceRat AI

A low-cost rodent-inspired micro-robot for early crop stress monitoring in smallholder greenhouses and high-bed farms.

RiceRat AI runs close to the ground to collect crop images and microclimate data, then uses AI to warn growers early about dry zones and stressed plants so small farms can act faster with lower-cost technology.

Explore the System See 3-Month Pilot

20-25 cm Compact rover length

3 features Focused pilot scope

1,000+ People reached through outreach

⟩

Quick Start Demo

⟩

Target Audience / Beneficiaries

RiceRat AI is designed to support smallholder farmers and farming communities that operate in space-constrained agricultural environments where conventional smart farming solutions are often impractical or unaffordable. The primary beneficiaries include:

Smallholder growers cultivating vegetables, strawberries, tomatoes, peppers, melons, and herbs in greenhouses or high-bed farms
Small agricultural cooperatives that manage multiple small plots but lack access to advanced precision agriculture tools
Educational farms, agricultural training centers, and university experimental gardens that can use the system both for crop monitoring and AI literacy activities
Youth and STEM communities who can engage with the project as a practical example of responsible, accessible AI for agriculture

These groups face a difficult trade-off: they need timely crop monitoring to reduce losses and improve productivity, but they cannot easily afford drones, dense sensor networks, or complex farm management systems. RiceRat AI is aimed at users who need a practical, low-cost, and understandable tool rather than a highly expensive automation platform.

Smallholder growers in greenhouses and high-bed farms

Agricultural cooperatives managing multiple small plots

Educational farms and university experimental gardens

Youth and STEM communities engaging with AI for agriculture

Key beneficiary groups that RiceRat AI is designed to serve — from smallholder farms to youth-driven STEM initiatives.

⟩

What RiceRat AI Does

Mobile moisture scouting

Stops at set points and builds a relative dry-versus-moist map.

Crop stress alerts

Flags yellowing, curling, and uneven growth from low-angle field images.

Simple grower dashboard

Shows which zone is driest, which needs checking first, and which is stable.

Problem

Manual scouting is slow, uneven, and easy to miss.

Smallholder growers in Vietnam still rely heavily on manual observation to detect water stress, early growth issues, and leaf abnormalities. Drones struggle in narrow greenhouse spaces, while fixed sensors only measure a few points and miss microclimate differences between beds.

Adoption

Users adopt it because it saves walks and sharpens decisions.

The value is not a cute robot. The value is one affordable device that patrols tight spaces, gathers repeatable data, and points growers to the beds that need irrigation or a closer look before the problem spreads.

Target Beneficiaries

Built for sites that need precision without enterprise budgets.

Smallholder growers

Vegetable, melon, tomato, pepper, and strawberry farmers in greenhouses or high-bed plots who need frequent crop checks but cannot justify drones or dense sensor grids.

Cooperatives and demo farms

Small cooperatives, education farms, and nurseries that benefit from repeatable data collection across multiple beds with minimal staffing.

University trial sites

Experimental gardens and teaching farms where embodied AI, sensing, and predictive analytics can be piloted in a realistic but manageable environment.

1. Target Audience / Beneficiaries

Who RiceRat AI is designed for.

Smallholder growers cultivating vegetables, strawberries, tomatoes, peppers, melons, and herbs in greenhouses or high-bed farms
Small agricultural cooperatives that manage multiple small plots but lack access to advanced precision agriculture tools
Educational farms, agricultural training centers, and university experimental gardens that can use the system both for crop monitoring and AI literacy activities
Youth and STEM communities who can engage with the project as a practical example of responsible, accessible AI for agriculture

Smallholder farming environments that reflect RiceRat AI's target users, especially growers working in bed-based and narrow-row cultivation settings.

Solution Design

A compact field robot with practical sensing and lightweight intelligence.

Hardware

20-25 cm body length, 10-12 cm width, and 8-10 cm height
4-wheel or mini-track chassis for greenhouse rows and high-bed edges
Front RGB camera for low-angle crop imaging
Air temperature, humidity, and light sensing for microclimate context
Soil moisture probe at stop points
ESP32 or Raspberry Pi controller with rechargeable battery and Wi-Fi or Bluetooth
Tail that works as an antenna detail or rescue tether

Software

Route-based image and sensor capture tied to specific beds
Simple dashboard for daily greenhouse review
AI scoring for water stress and crop anomaly warnings
Repeatable historical logs so the system can learn across time
Core outputs: dry zones, mild stress clusters, and row-level risk maps

Core Features

Three features only, so the pilot stays believable and useful.

Mobile moisture scouting

The robot pauses at defined points to measure soil moisture and estimate relative dry or moist conditions across beds.

Crop stress alert

RGB images plus environmental data help flag early signs such as mild yellowing, leaf curl, or uneven growth before they become obvious to the eye.

Daily route dashboard

The interface answers three grower questions quickly: which zone is driest today, which should be checked first, and which looks stable.

AI Approach

Lightweight models that are easier to train, explain, and deploy.

Computer vision

MobileNet or a small YOLO variant can detect basic visual abnormalities from RGB images without requiring heavy infrastructure.

Prediction

Random Forest or XGBoost can estimate irrigation need or crop stress from moisture, temperature, humidity, and compact image features.

Anomaly detection

Isolation Forest or learned thresholding can flag zones that deviate from recent normal patterns, even with limited training data.

Bootstrap with public data

Use PlantVillage, rice leaf disease collections, CCMT pest and disease images, and soil moisture or irrigation datasets to get the first baseline models running.

Finish with local pilot data

The real model must be refined on RiceRat-specific field images, row-level sensor logs, and simple labels such as dry, stable, or needs inspection from the pilot site.

2. Problem Statement

Why smallholders need affordable, fine-grained monitoring.

RiceRat AI addresses Google's focus area of Stronger Communities, with a secondary contribution to Scientific Progress through the practical use of embodied AI, mobile sensing, and lightweight predictive analytics for agriculture.

This challenge is highly relevant in Vietnam and across Southeast Asia because agriculture remains dominated by small-scale producers. In the Asia-Pacific region, family farmers produce about 80% of the region's food. In Vietnam, small-scale farms under 0.5 hectares make up nearly 70% of agricultural landholdings, and these smaller producers tend to have the lowest access to digital infrastructure and services ^[1].

Adoption of smart agriculture remains uneven, especially where farms are small, resources are limited, and technology must prove immediate practical value ^{[2, 3]}. Lam Dong, one of Vietnam's most important high-tech agriculture hubs, had 5,688 hectares of greenhouse area by Q1 2024, yet farmers often still rely on manual crop checking.

Drones are impractical in narrow or covered spaces, while fixed sensors only capture a few points and miss fine-grained variation across rows. The challenge is not a lack of agricultural data, but a lack of affordable, fine-grained, and actionable monitoring in small, space-constrained environments. This is precisely the gap RiceRat AI fills.

RiceRat AI system diagram showing Moisture Scout, Crop Stress Alert, and Risk Map Dashboard

AI System Diagram — RiceRat AI integrates Moisture Scout, Crop Stress Alert, and a Risk Map Dashboard to deliver zone-based early warnings from repeated route-based scouting.

Feasibility And Roadmap

A three-month pilot that stays ambitious but realistic.

Month 1

Robot setup and data collection

Design the chassis, integrate the camera and baseline sensors, operate it manually or semi-autonomously, and capture first image and moisture data from the pilot site.

Month 2

Dashboard and model baselines

Build the simple monitoring dashboard, train baseline irrigation and stress models, and test repeatable routes in one greenhouse or small garden environment.

Month 3

Pilot evaluation and tuning

Compare RiceRat outputs with manual checks, measure false alerts, and improve the warning logic before broader deployment.

Pilot In Vietnam

Start where greenhouse management and education pilots already make sense.

Da Lat greenhouses

High-value vegetable and strawberry production offers a strong environment for testing bed-level variation, moisture decisions, and repeated routes.

University experimental gardens

Controlled academic sites are ideal for quick data collection, labeling, iteration, and STEM engagement alongside the technical pilot.

Nurseries near HCMC, Binh Duong, and Lam Dong

Smaller commercial or education farms can validate whether the system remains useful outside a single showcase greenhouse.

Scalability And Impact

From one greenhouse pilot to a practical sensing platform for many small sites.

Scale pathway

Expand from one greenhouse to multiple beds in the same site
Move from demo farms to small cooperatives
Adapt from vegetables to other high-value greenhouse and net-house crops

KPI proposal

Reduction in daily manual scouting time
Increase in measurement points compared with current practice
Early detection rate for dry or abnormal zones
Number of households or sites onboarded after the pilot

Outreach to 1,000 people

4 university and STEM club workshops reaching about 400 people
3 demos for growers, cooperatives, and education farms reaching about 300 people
Short videos, infographics, and social content reaching 300 or more engaged viewers

References

Supporting evidence and context.

Burra, D., Hildebrand, J., Giles, J., Nguyen, T., Hasiner, E., Schroeder, K., Treguer, D., Juergenliemk, A., Horst, A., Jarvis, A., & Kropff, W. (2021). Digital agriculture profile: Viet Nam. FAO.
Dao, T. A., & Pham, C. N. (2022). Smart agriculture for small farms in Vietnam: Opportunities, challenges and policy solutions. FFTC Journal of Agricultural Policy, 3, 36–45.
Pham Thi Thanh Huyen, & Tran Tuan Anh. (2025). Digital transformation in agriculture in Vietnam: Trends, opportunities, and challenges. HUIT Journal of Science, 25(S3_ICA), 278–289.

⟩

First Two Confirmed Customers

Customer 01

Pineapple Farm in Dong Thap, Vietnam

Our first customer came from a pineapple-growing operation in Dong Thap. This was the first real-world validation that the RiceRat route-and-alert workflow solves an actual daily monitoring need on the ground.

Location: 92G7+G64 Hoi Cu, Dong Thap, Vietnam
Coordinates: 10deg22'34.6"N 106deg00'47.1"E

Open Pineapple Farm on Maps

Customer 02

Water Spinach Farm in Dong Thap, Vietnam

The second confirmed customer is a water spinach farm in the same province, giving us a strong next step to test consistency across crop types and refine early-stress signals in a new field context.

Location: 92G7+HMW Hoi Cu, Dong Thap, Vietnam

Open Water Spinach Farm on Maps

RiceRat AI

Not another generic farm robot.

Mobile moisture scouting

Crop stress alerts

Simple grower dashboard

Quick Start Demo

Target Audience / Beneficiaries

What RiceRat AI Does

Mobile moisture scouting

Crop stress alerts

Simple grower dashboard

Manual scouting is slow, uneven, and easy to miss.

Users adopt it because it saves walks and sharpens decisions.

Built for sites that need precision without enterprise budgets.

Smallholder growers

Cooperatives and demo farms

University trial sites

Who RiceRat AI is designed for.

A compact field robot with practical sensing and lightweight intelligence.

Three features only, so the pilot stays believable and useful.

Mobile moisture scouting

Crop stress alert

Daily route dashboard

Lightweight models that are easier to train, explain, and deploy.

Computer vision

Prediction

Anomaly detection

Why smallholders need affordable, fine-grained monitoring.

A three-month pilot that stays ambitious but realistic.

Robot setup and data collection

Dashboard and model baselines

Pilot evaluation and tuning

Start where greenhouse management and education pilots already make sense.

Da Lat greenhouses

University experimental gardens

Nurseries near HCMC, Binh Duong, and Lam Dong

From one greenhouse pilot to a practical sensing platform for many small sites.

Scale pathway

KPI proposal

Outreach to 1,000 people

Supporting evidence and context.

First Two Confirmed Customers

Pineapple Farm in Dong Thap, Vietnam

Water Spinach Farm in Dong Thap, Vietnam