RipeSense

Inspiration

In The Martian, Mark Watney transforms an unlikely environment into a place where life can grow. But growing food is only half the challenge. Knowing when to harvest, eat and preserve it is just as important.

On Mars, every resource matters. On Earth, we face the same challenge on a much larger scale: supermarkets, farms, delivery platforms and juice bars throw away huge amounts of fresh produce simply because nobody caught its ripeness at the right time.

That inspired us to create RipeSense, a smarter way to answer one simple question:

Should this fruit be eaten now, in three days, or not at all?

What It Does

RipeSense is an automated fruit-ripeness system: a robotic arm on a rail that picks fruit and runs it through a multi-sensor scanner. It was designed with Martian greenhouses in mind, but it is built for the places that waste the most fruit here on Earth.

The rail: the arm travels along a linear rail, past rows of fruit on a shelf, crate line or greenhouse bench.
The pick: the robotic arm picks up each fruit and places it in the scanner.
The scan: a depth camera, a spectral sensor and a gas sensor read the fruit at the same time.
The verdict: RipeSense labels it unripe, ripe, overripe or spoiled, estimates the days left, and the arm sorts it accordingly.

Our system aims to:

Detect ripeness: identify four stages of ripeness from colour, light and gas.
Estimate freshness: predict how many days a fruit has left.
Automate the sorting: no one has to squeeze every fruit by hand. The arm picks, scans and sorts.
Reduce food waste: flag fruit to sell, discount or use before it is lost.
Creative recommendations: ripe fruit gets a Gemini-generated recipe as a QR code, so a soft banana becomes banana bread instead of trash.

Where it fits:

Supermarkets and grocers: decide what goes on the shelf, what gets discounted today, and what goes to food banks.
Farms and packing houses: grade crates automatically and send the ripest fruit to the nearest market first.
Food delivery platforms: check fruit before it ships so customers never receive bruised produce.
Juice bars and restaurants: use fruit at its peak and turn overripe stock into smoothies.
Greenhouses, on Earth and on Mars: track a crop day by day, where every fruit counts.

The scanner also works without the arm: on its own with a slide-out drawer, it becomes a countertop scanner for a store front.

How We Built It

We combined robotics, computer vision, multi-sensor sensing, machine learning and custom hardware into one system.

Robotics: a UFACTORY robotic arm mounted on a linear rail moves between fruit, picks each one and places it in the scanner, then sorts it by result.
Sensing: an Intel RealSense depth camera for colour, bruises and size; an 11-channel spectral sensor that tracks chlorophyll fading; and a gas sensor that picks up the gases fruit gives off as it ripens and rots.
Machine learning: an image classifier trained on 20 public fruit datasets, plus a model that combines all three sensors into one verdict and keeps working if one sensor fails.
Hardware: a Raspberry Pi 5 with a touchscreen, inside a laser-cut plywood enclosure held together with 3D-printed joints and no glue.
Software: a touchscreen app that shows each result, keeps a scan history, and generates recipes with the Gemini API.

One of our main priorities was a flexible system that could adapt from everyday retail to future space agriculture.

Challenges We Ran Into

Building RipeSense involved much more than training an AI model.

Ripening is complicated. Different fruits change in appearance, firmness and gas output in different ways, which makes consistent predictions difficult.
Training data was messy. Public datasets use different labels, image conditions and definitions of ripeness, so we had to map everything onto our four stages.
Picking fruit gently. The arm had to grip fruit firmly enough to move it, without bruising it, and place it precisely in the scanner every time.
Making the arm, rail and scanner work together. Timing the rail movement, the pick and the scan into one smooth cycle took several iterations.
Hardware integration. Fitting multiple sensors into one sealed, dark scanner took several design rounds.

We also had to balance accuracy, processing power, physical constraints and a short timeframe. Each challenge pushed us to rethink our assumptions and improve our approach.

What We Learned

RipeSense taught us that solving a real-world problem takes much more than writing code or assembling hardware.

We learned how important it is to combine robotics, sensing, software and mechanical design, to question our first assumptions, and to understand the limits of the technology we use.

Most importantly, we realized that the most useful technology doesn't just collect information; it acts on it. A scanner tells you a fruit is ripe. A robot that picks it, scans it and sorts it actually saves it.

What started as an idea inspired by science fiction became a project with real applications on Earth.

What's Next

Piloting RipeSense in a real grocery store, farm or juice bar to measure how much waste it saves.
Faster picking and scanning for full-speed packing lines.
Improving prediction reliability across more fruit varieties.
Food-safe materials and a gentler gripper for delicate fruit.
Investigating how automated food monitoring could support future space missions.