Overview
Investigate how agents can interpret indoor spatial evidence to estimate location.
How it works
Given a floor plan and known anchors, MapAgent combines BLE, motion and map evidence to form constrained location candidates. The language-model agent selects a registered candidate or abstains. An independent verifier checks the proposed choice; accepted choices proceed, while abstention or rejection follows a deterministic fallback. Map reachability and motion consistency constrain the candidate set.
The agent chooses a registered candidate or abstains. An independent verifier accepts a valid choice; abstention and rejected choices use a deterministic fallback.
AI-assisted method illustration
BLE anchors, motion observations and a supplied map contribute complementary evidence. The floor plan and anchor positions are known inputs, not outputs inferred from scratch.
AI-assisted method illustration
A conceptual floor plan illustrates how walls, connected walkable space and motion consistency restrict location candidates. It does not reproduce any evaluation site.
AI-assisted method illustrationSelected contributions
- Developed a localisation agent using BLE, motion and map information.
- Evaluated localisation performance across six sites and 41 trajectories.
Research status
Ongoing final year project on indoor localisation and spatial reasoning.