Project archive.

MAP Rover was built for the point where the floor markings run out. It needed to know roughly where it was outside, not just show off individual features in a demo.

SLAM and an EKF helped estimate where the rover was and which data to trust next. A web interface kept that information handy without hauling the whole thing back to the bench.

GPS + local sensors
  → estimate pose
  → detect + plan
  → drive + report

MARTHA gave people a clearer view of crowding before they had to react. It paired hotspot awareness with lower-density options around campus, without making a walk to class feel dramatic.

The idea was simple: show where pressure was building, explain why it mattered, and make a lower-density option easier to choose.

campus activity
  → identify hotspot
  → explain context
  → suggest lower-density route

SAFE started with a simple idea: the lab is where data behaves. Homes are where the system needed to work. An app, a model, and an environmental sensor pod flagged temperature stress early.

The goal was not just a classifier. It was a useful heads-up: gather context, estimate risk, and make the next step clear.

signal + context
  → estimate risk
  → surface confidence
  → prompt next action

The India pilot and related ANN research tested whether the system could stay useful when data quality, living conditions, and daily routines did not resemble a clean dataset.

SAGE used a graph to keep related events together. Top-read Wikipedia articles became nodes, and weighted edges showed how their content connected.

IBM Watson sentiment analysis added another layer, so the network could be read through both its structure and the language around an event.

article + event
  → weighted edge
  → sentiment signal
  → graph view

SOAR put staff break-duty allocation and daily management in one place. The shared view replaced a chain of manual handoffs, which made the day a little less chaotic for everyone.

Head office and more than 200 staff at the International School of The Hague adopted the system. That meant day-to-day usability mattered as much as the data model.

staff + duty constraints
  → allocate break cover
  → publish roster
  → resolve exceptions