AV on trial in Punggol nearly hits girl; Grab directed to narrow testing as LTA probes near miss

An autonomous vehicle, or AV, is a vehicle designed to drive using onboard systems that sense the road, traffic, and people around it. It can perform driving tasks without a person continuously controlling the steering or pedals. This matters because AVs must respond safely to unpredictable road situations.

A safety operator is a human inside the vehicle who supervises the test and can intervene. In Punggol, the AV detected and tracked two pedestrians. When the girl began crossing, the vehicle started braking and steering away. The safety operator then took over, continued braking, and steered farther away from the pedestrian.

The incident shows why human supervision remains important during public-road testing. The operator’s intervention helped the vehicle avoid the girl, but LTA still directed Grab to pause testing at such junctions while it investigates. Grab is reviewing the case with LTA and WeRide.

At a Punggol junction, a Grab AV was turning left after its traffic light turned green. A girl stepped off the kerb and began crossing while the pedestrian signal was also green. The vehicle did not stop before the crossing, and the situation became a near miss.

The girl moved into the vehicle’s path, then halted as the AV passed in front of her. The AV appeared to slow down, began braking, and steered to the right. A woman accompanying the girl pulled her away from the vehicle. The safety operator then took over, continued braking, and steered away from the pedestrian.

LTA said preliminary data showed the AV had detected and tracked both pedestrians. It is investigating the response. The video, posted on Facebook on Nov 5, attracted more than 260,000 views. LTA directed Grab to pause testing at such junctions.

Grab’s AV operation in Punggol was already a public service, not only a private experiment. Its shuttle services had been available to the public since April 1. This means residents could use AV shuttles on established routes before LTA ordered a targeted testing pause.

The operation covered two shuttle routes. One connected Punggol Matilda Court with Oasis Terraces. The other connected Punggol Matilda Court with Punggol Coast Mall. Grab had also announced a trial allowing people to book rides between specific pickup and drop-off points along its existing routes.

The near miss happened while an AV was testing a new route. That route was being prepared for enhanced on-demand travel within Punggol. The service was expected to expand to the wider public in the fourth quarter of 2026, but the junction pause affects testing in some scenarios while LTA investigates.

LTA’s direction means Grab must pause AV testing at certain junctions while investigators examine the near miss. The pause is targeted at such junctions, rather than described as a shutdown of every AV service in Punggol. LTA will also review whether the vehicle’s response to similar scenarios should be adjusted.

The affected testing supports enhanced on-demand travel within Punggol. That service is expected to expand to the wider public in the fourth quarter of 2026. Grab had been preparing by testing a new route and had announced plans to trial bookings between specific points on existing routes.

The immediate effect is likely to be a delay or change in testing at junctions with similar turning and crossing situations. However, the article does not say that the wider expansion has been cancelled. Its timing and design may depend on LTA’s investigation and any required safety changes.

Traffic signals can give a turning vehicle and a pedestrian a green light simultaneously because both movements are permitted, but their paths may cross. A green signal does not always mean every possible movement is conflict-free. Turning drivers or vehicles must account for pedestrians using the crossing.

At the Punggol junction, the AV received a green light to turn left, while the pedestrians also had a green signal. LTA said the pedestrians were stationary for some time before the girl began crossing. The vehicle then started braking and steered away, while the safety operator took control.

The article does not set out a full traffic rulebook. It does say LTA reminded pedestrians to watch for oncoming traffic and wait for it to completely stop before crossing. In practice, safe avoidance depends on both road users checking the shared space and yielding when their paths conflict.

Autonomous vehicles detect pedestrians through onboard sensing systems and software that identify objects, track their positions, and estimate how they may move next. The article confirms that the Grab AV detected the two pedestrians and continuously tracked their movements. It does not describe the vehicle’s specific sensors or prediction software.

In this case, the pedestrians were initially stationary, then the girl began crossing as the AV turned left. The vehicle started braking and steered right, while the safety operator took over and continued both actions. This shows the key mechanism: noticing a person is only one step. The vehicle must interpret changing movement and respond safely.

A near miss can still occur if a pedestrian changes direction, enters the path late, or leaves too little time for braking. LTA is investigating whether the AV’s response to such scenarios should be adjusted. This explanation uses established AV principles beyond the article’s technical detail.

Safer AV testing can begin away from busy public roads. Companies can use computer simulations, closed test tracks, and controlled junctions to examine turning, braking, and pedestrian scenarios. These settings allow repeated testing without exposing ordinary road users to every early-stage failure.

A company could then run a small, geofenced pilot with low speeds, fixed routes, trained safety operators, and close monitoring. Authorities could require testing during selected hours, with extra observers and clear stop conditions. Scenarios involving children, sudden crossings, and simultaneous green signals could be rehearsed before public expansion.

The article does not list these alternatives. They are established approaches for managing AV trials. In this case, LTA has already required Grab to pause testing at similar junctions while investigating. The findings could guide changes to the vehicle’s response and determine when testing can safely resume or expand.