Understanding the scene. Anticipating the event.

01 / The challenge

The question behind the system.

An isolated detection does not explain what may happen next. The research question is how to combine observations, relationships and learned dynamics to interpret a scene and anticipate critical events in an aviation setting.

This research explores how an AI system can move beyond detecting individual objects to reason about a changing scene. Multimodal perception, semantic graph reasoning and model-based reinforcement learning are brought together in a Unity 3D aircraft-landing scenario to investigate runway risks and decision support.

02 / The approach

From architecture to implementation.

1. 01

   Combine multimodal observations with a semantic representation of the scene and its relationships.

2. 02

   Use graph-based contextual reasoning and model-based reinforcement learning within a hybrid architecture.

3. 03

   Explore the architecture through a Unity 3D aircraft-landing case study and document the research in a conference paper.

03 / Outcomes

What the work shows.

• The project presents a simulation-based case study for scene understanding and critical-event prediction.
• The resulting paper was published at RoboticCC 2025 and is available through IEEE Xplore.