What is happening at the lab.

The whole group, three days, and one question: what do we want to have shown by this time next year. We agreed the shared evaluation setting across the prediction and planning work, decided which datasets we maintain and which we retire, and planned the next recording campaign. The evening programme settled nothing scientific and improved everything else.

Six teams, one taped course, and rubber ducks as the obstacles that decide the ranking. Every car drove the course autonomously by the end of the afternoon, which is not something we could promise at the start of the semester. Several of the participants are now writing their theses with us.

Visitors of every age climbed around the vehicle from early evening until late. The sensor rig on the roof was the thing people asked about most — how far it sees, what happens in rain, whether it recognises children. A good reminder that the questions the public asks are the ones the field has not fully answered.

An afternoon of demonstrations and discussion around the simulator and the research bicycle, with a walk through the current work on VRU perception and planning.

We showed the full stack on the vehicle: VRU perception, prediction and the planner’s decision, live on the test site. The instrumented bicycle drew the most questions — several partners had not seen cyclist action recognition running outside a paper before. Between the runs the tent was busy from morning to evening.

A closed airfield lets us stage the situations that are too dangerous to wait for in traffic — someone stepping out from behind a parked car, a cyclist cutting across, a crossing at dusk. The week produced the recordings behind LIF-Net and bikeActions, and a long list of calibration lessons for the next campaign.