We are studying how building connections, reported conditions and incomplete observations affect candidate routes. Here are the ideas guiding the experiments, a small demonstration, and the limits of what each can establish.
Valkor is in development. Hardware checks and synthetic routing experiments are separate workstreams. Dynamic emergency routing is not deployed as an active school safety system.
How the work connects
Observations in. Building context. Reviewable routes.
Valkor brings together two connected development efforts: indoor location hardware that supplies observations, and a routing engine that evaluates a building model and reported conditions. Their integration is a development milestone; the routing experiments currently use simulated inputs.
01
Location hardware supplies observations
Tags and antennas provide observations about a device’s presence or direction. An observation needs a source and timestamp, with quality and confidence where available. A missing signal leaves a gap in the evidence; a badge detection alone does not prove who is present or whether a doorway is usable.
02
The engine evaluates a building model
Rooms, junctions and exits form a graph connected by doors and corridors. The engine combines that model with reported closures and available observations to find candidate paths. A closure must be supplied by a defined source; location hardware does not automatically detect hazards or blocked doors.
03
People review the result and its limits
The output should explain the route, the inputs used, and whether the evidence supports a result. Known closures can make an exit unreachable; stale or conflicting reports can leave insufficient information. Any future operational use would require validation and established staff and responder procedures.
A connected path is a statement about the model and its inputs. It does not establish that a route is safe, accessible to a particular person, or suitable for an emergency.
Research informing the experiments
Changing paths. Access to exits. Movement through doors.
These papers help frame questions and comparison methods. The applications to Valkor below are our proposed adaptations; the cited authors have not evaluated or endorsed Valkor.
Changing networks
Evolving network representation learning based on random walks
EvoNRL studies how to update a learned representation when a network changes. It repairs affected random walks and uses changes in those walks to decide when to refresh the representation.
How it informs our experiments
For Valkor, this raises a useful question: after a doorway closes, what needs to be recomputed? Our immediate experiments use explicit paths and closure checks. Learned representations remain a later research direction.
The paper evaluates network representations. It does not validate emergency routing or establish which building route is safe.
Scoring and classifying regions via multimodal transportation networks
Aaron Bramson, Megumi Hori, Bingran Zha and Hirohisa Inamoto · Applied Network Science 4, 97 (2019)
This study combines walking, road, bus and train networks to compare what different locations can reach within a travel time. It shows why connectivity alone does not describe how useful a connection is.
How it informs our experiments
We adapt that question to a building: which rooms retain access to an exit after a connection is removed, and how do the remaining paths change? We assess reachability separately from distance and modeled travel time.
This is transportation accessibility research. It does not establish disability access, building code compliance or school evacuation performance.
New insights into pedestrian flow through bottlenecks
Armin Seyfried, Tobias Rupprecht, Oliver Passon, Bernhard Steffen, Wolfram Klingsch and Maik Boltes · Author manuscript (2007)
Laboratory experiments examine how people move through bottlenecks of different widths, including flow, density and time gaps between people. The results challenge simple assumptions about when queues will form.
How it informs our experiments
This informs our plan to compare simulated doorway crossings, queues and completion against documented measurements. Matched comparisons keep the starting population and seed fixed while changing one condition, such as a doorway closure.
A path can exist while movement still stalls. An unfinished run has no complete clearance time, and a software benchmark does not establish real school performance.
Change a reported condition and inspect the result. This simplified browser model demonstrates the research prototype’s route availability policy. It runs a shortest path calculation over the illustrated graph.
SIMULATED DATA · Fictional building, assumed lengths and synthetic reports. No live hardware or school data. This is a graph demonstration, not a pedestrian simulation or emergency instruction.
Orange: candidate pathRed dashes: closedYellow dashes: stale reportAll connections are bidirectional
Computed outcome
Candidate path found
FOUND
All four connections have simulated fresh OPEN reports.
Classroom → Hall junction → West exit
Modeled path length: 14 m. The shorter west path is available. Only connections with fresh OPEN reports are eligible.
This demonstration does not model crowds, hazards, individual mobility needs or travel time. A found path is not a safety assessment.
How the result is calculated
The browser uses Dijkstra’s shortest path algorithm with assumed lengths of 8, 6, 12 and 10 metres. It searches fresh OPEN connections first. If no path is found, a second search allows stale connections while retaining known closures. A possible path in that second search produces INSUFFICIENT_INFORMATION; no possible path produces UNREACHABLE. It does not display the uncertain path as a recommendation.
The baseline west path is 14 m. Closing the west doorway leaves a 30 m east path. The origin, destinations, graph and lengths stay fixed across all four conditions. This is a separate, simplified implementation of the prototype’s graph policy, not a replay of a measured or full simulator run.
We welcome specific critiques, technical contributions and conversations about a controlled evaluation. Start with one question, the experience you bring, and the evidence that would make an answer useful.
Researchers
Review a routing assumption, suggest a reproducible comparison, or help define how to measure doorway flow and uncertainty. A useful starting contribution is a written critique of one experiment and the evidence it would need.
Help investigate a specific failure case, review graph and observation interfaces, or improve reproducible tests. Start with a small scenario, expected behavior and a way to check the result.
Describe a real information gap, walk through the current process, and identify what staff would need to verify. You can also discuss a limited evaluation with agreed devices, zones, hours, privacy rules and stop conditions.