A route that was usable yesterday can become unsuitable during the next shift. For construction teams considering mobile robots, connecting movement to the work programme is a practical challenge. An emerging navigation study offers a useful case for examining that connection and the evidence needed before a trial.
What CORNAV adds
CORNAV, first posted on 2 October 2026, combines 2D drawings, scene observations and project schedules to plan robot routes without requiring BIM. Room boundaries link instructions to locations; schedules activate restrictions, while a language-model check can escalate hazards before planning. Its accessible publication is an arXiv preprint. [1]
Read the test results in context
In a 54-request comparison, the reported task-success score was 72.2%, versus 13.0% with blueprint information removed and 7.4% for the semantic-navigation baseline. Numerical evaluations used recorded Go2 survey data from office and construction environments. Live Go2 and G1 demonstrations were reported separately. Reported exclusion-zone compliance describes the authors’ finite evaluation, rather than certified safety performance. [1]
For a project team, these distinctions shape the questions to ask. A planning score should be accompanied by information about the test environment, the requests counted and the cases rejected. Ask what happened when a destination was inaccessible, how a person intervened and which failure conditions were outside the evaluation. A headline percentage alone cannot answer those operational questions.
A planned route and movement control are separate checks
Established navigation software illustrates this distinction. Nav2’s official keepout-zone tutorial describes restrictions in both the global map used for route planning and the local map used during movement. A restriction applied only to planning does not automatically provide the same local control behaviour. This is background context for assessing navigation systems, not a claim that this tutorial validates CORNAV. [2]
As a hypothetical trial question, consider a corridor closed for overhead work. The team could ask the supplier to demonstrate how an authorised closure reaches the robot, how its effective time is represented and how receipt is confirmed. It should also be clear who can reopen the route and what the system does when an update is missing. These are evaluation questions, rather than operating instructions for a particular robot.
October guidance highlights the changing-worksite problem
A NIOSH Science Bulletin published on 7 October 2026 identifies changing work areas, overlapping trades and environmental interference with sensing as important construction-robotics issues. It also highlights communication failures, software changes and unexpected behaviour after faults. The bulletin recommends considering risks during design and task assessment, alongside continued evaluation of worker–robot interactions as conditions change. [3]
That guidance helps explain why navigation accuracy is only part of readiness. Dust, lighting, visibility and unfamiliar workers entering a robot’s workspace can affect an installation. NIOSH’s discussion concerns the United States and provides research and safety context; it is not an Australian approval pathway. [3]
How Australian teams can frame a trial
Safe Work Australia’s plant-risk model code covers people who manage or control plant and those who install or commission it. The organisation advises checking whether the code has legal effect locally. Its current page identifies a November 2024 update, so this is established Australian background rather than a new October rule. [4]
Civil Insight’s assessment is to start with a bounded use case and explicit acceptance evidence. A trial proposal could identify its permitted area, supervision arrangements, drawing and programme revisions, update responsibilities and review points. Keep records of rejected missions and interventions alongside successful ones. This makes the limitations visible to the people deciding whether to expand the trial.
The useful development to watch is the connection between project information and robot behaviour. For adoption decisions, request evidence linking the proposed workflow to the actual environment, then assess its results against the project’s requirements.
Sources and research status
[1] Ali Pour, P. et al. (2026). CORNAV: Construction-Aware Reasoning for Robot Navigation on Active Worksites. Preprint, 2 October 2026. DOI: 10.48550/arXiv.2610.03622. Full text. No peer-reviewed venue is listed in the record checked.
[2] Nav2 contributors. Navigating with Keepout Zones, official documentation. Technical background, accessed 27 October 2026.
[3] Castleberry, A. et al., NIOSH. Safety in Construction Robotics, 7 October 2026. US research and safety context.
[4] Safe Work Australia. Managing risks of plant in the workplace, model code page, updated 14 November 2024. Australian background.
Sources checked 27 October 2026. Cover photograph: Unitree Go2 at a public parade in Cluj-Napoca, Romania, 24 May 2025, by Lucian Nuță, via Wikimedia Commons, CC BY-SA 2.0. Resized for display. Historical illustration of the robot model, separate from the CORNAV experiments.
.jpg){kind=link}