Unnumberium
The physical learning landscape

From reading answers
to understanding actions.

AI education is moving from screens into the physical world, from reading answers to understanding actions.

That is our reading of an emerging direction—not a claim that every system below already understands a learner’s thinking. Here is where the work overlaps, where it differs, and what the public evidence supports.

Sources reviewed · 21 October 2026

The Unnumberium lens

The action is part
of the answer.

Unnumberium is developing a voice-first, screen-free approach to arithmetic. Children work with real mathematical signs and objects while a camera observes their physical actions. The aim is to connect what a child does with what they say, so guidance can respond to their strategy as well as their result.

Voice first

Questions, explanations and guidance happen through spoken conversation.

Materials at the centre

The child’s activity stays with physical quantities and mathematical signs. A screen is not the centre of the learning task.

Follow the strategy

The goal is to interpret how a child counts, groups and rearranges materials over time—not just recognise a final configuration.

This describes our product direction, not an independently validated claim of complete strategy understanding or better learning outcomes. Physical perception and pedagogical interpretation remain capabilities that must be tested.

Seven perspectives

A field, not a single category.

Blockplay.ai and Lantern have close overlap in particular parts of the experience. Osmo, NoRILLA and CodeeBot are adjacent educational alternatives. CSU is research; Prism is an adjacent technology. These are editorial classifications, not a claim that all seven are direct competitors.

01

Close technical overlap · Research prototype

Blockplay.ai

What the sources document

The project documents a downloadable Windows prototype that recognises Cuisenaire rods and multilink cubes under a webcam, with spoken counts, colours, tones and animations.

Evidence boundary

This is developer documentation and a public demo, not an independent reliability test. Its examples ask users to remove their hands; they do not establish continuous interpretation of a child’s strategy.

Where it sits beside Unnumberium

A close reference for recognising mathematical manipulatives. Unnumberium’s intended focus extends to the sequence of mathematical actions and a spoken tutoring conversation.

Project, demo and downloads ↗

02

Close interaction overlap · Announced product

Lantern · Scout Learning Lab

What the sources document

The official site describes a camera-equipped tutor for books, reading lessons and homework, with spoken coaching and a parent companion app.

Evidence boundary

These are announced capabilities. The site says it is testing with families and plans to ship in 2027. Public documentation does not establish recognition of manipulative-based mathematical strategies.

Where it sits beside Unnumberium

Close to the screen-free, conversational tutoring experience. Its published emphasis is pages and homework; Unnumberium focuses on mathematical thinking expressed through physical materials.

Product and availability ↗ · Operator and privacy policy ↗

03

Adjacent product · Tangible tablet games

Osmo

What the sources document

Osmo documents physical game pieces, a tablet base and camera reflector. Tabletop activity feeds games on an iPad or compatible Fire tablet, including mathematics and coding.

Evidence boundary

The screen is part of the documented play loop. The reviewed source does not establish an open-ended voice tutor or general interpretation of a child’s mathematical strategy.

Where it sits beside Unnumberium

An established reference for connecting physical play to digital feedback. It is an adjacent alternative, rather than the same screen-free tutoring format.

How Osmo works ↗

04

Adjacent educational system · Research-backed deployments

NoRILLA

What the sources document

Carnegie Mellon documents intelligent science stations that observe physical experiments and give personalised feedback through an on-screen gorilla. Examples include building towers and testing them on a shaking table.

Evidence boundary

This is a documented application of computer vision to guided scientific inquiry. Its published studies concern specific tasks and settings; they do not prove general tutoring ability or superiority over Unnumberium.

Where it sits beside Unnumberium

Strong evidence that action-aware tutoring already exists. The difference is the learning format and focus: science stations with a virtual character versus Unnumberium’s voice-first physical mathematics.

CMU system overview ↗ · CMU study and learning design ↗

05

Research direction · Not a commercial competitor

Colorado State University

What the sources document

A 2026 EDM doctoral-consortium paper by Changsoo Jung and Nathaniel Blanchard investigates vision-language models as perception and judgment modules for physical block-construction tutoring.

Evidence boundary

The preliminary study reports a gap between identifying an action and judging spatial correctness. A full multimodal tutoring pipeline and learning-outcomes validation are proposed work, not completed results.

Where it sits beside Unnumberium

A relevant research challenge for the field: seeing a move is only one step toward understanding it. This is research context, not a purchasable alternative to Unnumberium.

EDM 2026 paper ↗

06

Adjacent technology · Spatial guidance and robotics

Prism · The Light Company

What the sources document

The company’s Prism vision page presents projected interfaces and piano guidance. Its current main site focuses on staging and visually verifying physical conditions for robot-policy evaluation.

Evidence boundary

The demos and deployment claims are company-published. They do not establish a child-focused mathematics tutor, pedagogical outcomes or interpretation of children’s mathematical strategies.

Where it sits beside Unnumberium

A useful example of computing moving onto physical surfaces. Projected guidance and robotics evaluation make it a technology analogue, not a direct education competitor.

Prism vision and demos ↗ · Current robotics focus ↗

07

Adjacent product · Tangible coding kit

CodeeBot

What the sources document

The official product page describes magnetic coding blocks, light-based execution feedback, robotics and GPT tutor guidance in a screen-free coding kit.

Evidence boundary

These are vendor claims. The reviewed pages still contain launch and preorder messaging; they do not by themselves verify fulfilment, tutoring effectiveness or camera-based observation of a child’s actions.

Where it sits beside Unnumberium

Relevant to hands-on coding and AI-assisted discovery. Unnumberium’s intended domain is arithmetic with physical mathematical materials, rather than programming a robot with coding blocks.

Product description ↗ · Availability messaging ↗

Learning belongs in the world.

Our starting point is mathematics you can touch.

Explore Unnumberium