Interactive 3D Tool Treellis Fosters Human Learning in Artificial Intelligence Design

A 3D modelling tool developed by researchers at the University of Sydney has demonstrated how people can achieve a deeper understanding and learn complex technical concepts while collaborating with artificial intelligence systems. Created by an engineering research team in partnership with the University of Cambridge and Singapore Management University, the open-source software, known as Treellis, was designed to address mounting concerns across the technology sector that individuals increasingly outsource their critical thinking to automated tools in ways that inhibit their own learning. While conventional generative platforms can swiftly convert short text prompts into finished digital models, that automated output typically conceals the underlying construction process, leaving users unaware of how the structures were assembled or how to modify them independently.

To explore whether co-creating with artificial intelligence could instead foster independent problem-solving skills, the researchers developed Treellis to guide users through iterative design assemblies rather than generating immediate, uneditable models. The system combines manual editing with automated assistance, allowing operators to progressively refine three-dimensional shapes. The lead author of the research paper, Yixiang Chen, a doctoral candidate in the Faculty of Engineering at the University of Sydney, noted that the platform supports transparent and collaborative co-creation because it shows users how objects are made rather than merely producing completed items for them. The tool is open source and freely accessible to anyone using a standard web browser, and developers can integrate its interface into third-party artificial intelligence platforms through an application programming interface.

Associate Professor Dr Anusha Withana, who led the development of Treellis within the School of Computer Science at the University of Sydney, explained that the software allows users to manipulate and control their work without relying entirely on automated algorithms. According to Dr Withana, permitting users to experiment and make mistakes during the assembly process forms an essential element of how people acquire practical design knowledge. The system was structured to ensure that individuals retain agency over each component, breaking away from standard workflows where automated systems function as opaque black boxes that hide their internal operations.

The research team evaluated the educational impact of the system in an experimental study involving human participants divided into two test cohorts. Each participant was tasked with building three digital objects of increasing structural complexity subject to specific design requirements: a drinking cup, a desk, and a computer monitor. The baseline comparison group relied on a standard prompt-based artificial intelligence tool to generate their models. In contrast, the second group used Treellis, which rendered every component part of each object and its corresponding spatial relationships completely visible on screen. Through this graphical interface, users could directly click on individual elements, progressively modify dimensions and positions, and immediately observe how those adjustments altered the overall model.

During the experimental trials, researchers observed clear behavioral differences between the two user groups. Participants working in Treellis actively investigated the underlying construction of their models and tested small, reversible adjustments until their work matched the assigned targets. This hands-on, progressive manipulation enabled them to grasp sophisticated three-dimensional design principles intuitively. In contrast, participants assigned to the prompt-only baseline system frequently failed to understand the structural assembly of their designs. When baseline users attempted to correct specific components by submitting revised verbal prompts, the automated software frequently disrupted other sections of the objects, introducing unexpected errors elsewhere across the models.

According to Dr Withana, the interactive workflow of Treellis directly resolves two fundamental barriers commonly encountered in human-computer interaction: the gulf of evaluation and the gulf of execution. The gulf of evaluation describes the difficulty a human user faces when trying to comprehend how a computational system operates or interprets input. The gulf of execution refers to the struggle an individual experiences when attempting to translate an intended objective into actionable commands within that software. By providing direct, transparent manipulation of individual components, the interactive functionality bridges both gaps simultaneously, preventing users from completely offloading cognitive tasks to automated algorithms while enabling them to understand and alter generated outputs.

The research team highlighted that the software could be expanded beyond conventional design settings to assist professionals who have little or no formal training in three-dimensional digital modelling. In particular, Dr Withana and his colleagues hope to introduce the open-source interface to healthcare practitioners in clinical environments. Medical professionals could employ Treellis to interactively build personalized assistive devices for patients with disabilities, including custom-fitted ankle-foot orthotics for children living with cerebral palsy, tailor-made wrist splints for individuals suffering from arthritis, and ergonomically tailored grips for daily rehabilitation equipment.

The study was documented in a conference paper titled ‘Treellis: Creating In-Situ Parametric Editing Tools to Progressively Design AI-Generated 3D Models’, which received a Best Paper Award at the Association for Computing Machinery Symposium on User Interface Software and Technology, held in Michigan. Best Paper honours at the symposium were granted to the top two percent of submissions, selected from a total of 251 research papers delivered at the gathering. Professor Anne Roudaut of the University of Bristol, who served as chair of the award committee, commended the project for its originality, scientific rigour, and potential to establish new directions for future human-computer interaction research. The authors declared no competing interests in connection with the study.

Subject of Research: Human-computer interaction and progressive learning in AI-assisted 3D design

Article Title: New tool shows how humans can learn with AI

Article References: New tool shows how humans can learn with AI. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, human-computer interaction, 3D modelling, Treellis, University of Sydney, open-source software, UIST 2026, collaborative design, tool, shows, humans, learn