Intelligent and Collaborative Decision Making for the Airspace of Today

We study human-centred AI for today's air traffic management — systems that surface tacit operational knowledge, support collaborative decisions, and turn human preferences into verifiable designs.

Our Research Prototypes Visit our Blog

The lab

For the ATM systems of today, not tomorrow.

Intuelle is an academic research lab led by Thinh Hoang at Van Lang University. We investigate how human-centred AI can support decision making in contemporary air traffic management, with a focus on tacit operational knowledge, human–AI collaboration, and optimization methods that reflect human preferences. Our research produces prototypes and methods that can be studied, challenged, and verified against the systems and constraints in use today.

Institution Van Lang University Faculty Automotive Engineering Department Aerospace Engineering

Research collaborations

Our work is strengthened through active collaboration with ATM and aviation research institutions across Europe, North America, and Asia.

  • EUROCONTROL Innovation Hub Brétigny-sur-Orge · France
  • IASL at George Washington University Washington, D.C. · USA
  • ENAC Toulouse · France
  • CAFUC Chengdu · China

What we work on

01

Decision-Making & Agentic Systems

Search, planning, and language-model agents that reason over real operational air-traffic decisions.

02

Transportation & Optimization

Large-scale flow regulation, sequencing, and routing, formulated as tractable optimization.

03

Verification & Trust

First-class tooling to inspect, replay, and verify that a proposed solution actually holds under the real scenario.

The systems

Our Research Prototypes

Each prototype is a research module within our broader work on human-centred decision making in ATM. We study how optimality and interpretability can reinforce one another, enabling humans and machines to collaborate through shared representations and a common operational picture.

SYS-01 Project Cortex

Flow's Kitchen

A verification-first ATFM workspace designed for compatibility with EUROCONTROL Network Manager systems.

Flow's Kitchen is a web-native research environment built around EUROCONTROL Network Manager workflows and operational data. It makes optimization results across an entire traffic day inspectable by translating raw ATFM scenarios and solver output into familiar operational concepts — flow groups, regulations, vulnerable traffic volumes, and slack — all replayable on a live map.

  • EUROCONTROL NM
  • ATFM
  • Visualization
  • Verification
  • Web-native
Open Flow's Kitchen
Flow's Kitchen: European ATFM map with flight list and flow-group regulation panel.
Flow's Kitchen — regulation view over the European network
Flow's Kitchen: overload heatmap of traffic volumes with a highlighted flow.
Traffic-volume overload, per flow group
SYS-02 Project Parrhesia

RegulationZero

Flow-centric regulation by Monte Carlo Tree Search.

A search-based planner that regulates traffic by flows rather than flight-by-flight. RegulationZero explores the space of regulation strategies with MCTS, consistently reducing overload while scaling to problems where flight-centric methods stall.

  • MCTS
  • Regulation
  • Optimization
  • Scale
Read our preprint
Monte Carlo search over regulation strategies
SYS-03 Project Tailwind

Flow's API

A Network Manager–compatible backend for scenarios and orchestration.

The data and orchestration layer beneath everything else. Flow's API serves traffic-scenario data through a Network Manager–compatible interface and coordinates optimization runs, so every tool and solver speaks one language.

  • Backend
  • Orchestration
  • NM-compatible
  • API
Open API Docs
One backend, many clients and solvers
SYS-04 Project Gemini

Flow's Predict

Learning routing preferences with inverse reinforcement learning.

Controllers and airspace users express preferences that never appear in a flight plan. Flow's Predict recovers those preferences from observed trajectories with inverse reinforcement learning, producing routing choices that match how the network is actually flown.

  • Inverse RL
  • Routing
  • Preference learning
Try in Flow's Kitchen
A recovered reward field over the route space
SYS-05 Project Rustlingtree

SequenceZero

An ADS-B playback and agent harness for terminal-area sequencing.

A browser-based environment for replaying real ADS-B traffic and building agentic workflows around the terminal-area sequencing problem. SequenceZero pairs a full playback timeline with conflict, feasibility, and separation tooling — and an in-context agent that proposes and checks resolutions.

  • ADS-B
  • TMA sequencing
  • Agents
  • Playback
SequenceZero conflict-check view over the Dallas–Fort Worth terminal area with a ranked conflict list.
Conflict check over the Dallas–Fort Worth TMA interface: ApproachClaw
SequenceZero feasibility check with an in-context agent reasoning about a speed advisory.
Feasibility check, with the in-context agent
SYS-06 Project Hail Mary

SequenceBearings

Causal learning for terminal-area procedure design.

An automated approach to designing terminal procedures. SequenceBearings learns causal heuristics for path stretches and speed advisories, searching for procedures that maximize throughput while minimizing risk.

  • Causal learning
  • Procedure design
  • Throughput / risk
A causal graph from levers to outcomes