Grab pauses Singapore driverless testing after pedestrian near-miss

Singapore’s Land Transport Authority ordered Grab on Thursday to suspend autonomous testing at similar Punggol junctions after a purple shuttle nearly struck a schoolgirl early that morning, while a full safety investigation began.

Dashboard footage circulating on social media showed the shuttle begin a discretionary left turn while the traffic light was green. A young schoolgirl stepped onto the pedestrian crossing at the same time. The vehicle braked abruptly to avoid collision, and the onboard safety operator took manual control of the steering wheel and guided it away from her.

The sequence triggered immediate and widespread concern about driverless shuttles sharing residential roads with children and other vulnerable road users. LTA’s order covers testing at comparable junctions; the authority did not announce a nationwide suspension of autonomous services.

Takeaway: A near-miss involving one child triggered a targeted testing suspension, not a national shutdown.

What will the safety review examine?

LTA said the shuttle "was testing a new route" for Grab’s planned on-demand passenger service. The agency will review how the system detected and responded to the pedestrian, then decide whether its artificial-intelligence systems need adjustments for complex crossings.

Grab and WeRide, the technology partner behind the shuttle’s driving system, are conducting a parallel review of the automated response and object-detection systems. Reviewers will reconstruct the turn, the girl’s movement and the safety operator’s intervention to establish how the automated system and human control performed.

Investigators will soon release a technical report detailing the sequence of events. Grab must satisfy LTA’s safety requirements before the regulator permits junction testing to resume, giving the decision weight beyond one route.

Takeaway: The technical report, rather than the widely shared video alone, will determine whether the system needs correction.

Why does Grab’s expansion increase the stakes?

Grab currently operates two fixed autonomous shuttle routes in Punggol using WeRide-developed driving technology. The company has announced plans to:

• Increase its fleet from 10 to 50 vehicles over six months, a fivefold rise that adds 40 vehicles.
• Launch on-demand travel across selected locations within Punggol.
• Test new routes for that service.

If the expansion proceeds, more automated vehicles will enter residential streets and complex pedestrian settings before the review concludes. Investigators have not identified a cause, so any fleet-wide effect remains unproven.

Grab representatives said passenger and pedestrian safety "remains the absolute top priority" in its autonomous operations. That pledge now faces a practical test: LTA must determine whether the near-miss was isolated or requires correction before wider deployment.

Takeaway: The planned fivefold fleet increase makes the investigation consequential for Grab’s timetable and Punggol residents.

How did the near-miss happen?

The recorded sequence shows three linked tasks: detect the girl, predict that she would enter the shuttle’s path, and stop or redirect the vehicle safely. The video makes the final movement visible, but it cannot establish which software function performed as intended.

Autonomous systems must handle unpredictable pedestrian movement, particularly during permitted turns at busy neighborhood intersections. Engineers therefore need to refine object-detection algorithms and ensure driving software acts before a collision becomes imminent, rather than only at the last braking moment.

The review will test the chain from camera-data processing to automated response and handover to the safety operator. Developers recognize that public trust depends on transparency about how these systems interpret visual data.

Takeaway: Investigators must test the entire perception-and-control chain before judging the operator’s intervention.

Who else is testing autonomous transport?

Residents have expressed unease about sharing neighborhood roads with artificial-intelligence systems that remain in beta testing. The video drew criticism from parents and local road-safety advocates. Urban transport experts have called for stricter operating rules before driverless services expand beyond Punggol into other major Singapore towns.

Grab does not operate alone. ComfortDelGro manages one driverless shuttle route with Pony.ai vehicles. Singapore’s government supports these trials to ease severe manpower shortages in the transport sector, while regulators balance that objective against the safety of vulnerable road users.

Industry leaders remain confident that machine learning will eventually make autonomous vehicles safer than distracted human drivers. The near-miss leaves that claim unproven, but LTA’s response will shape operating standards for Singapore’s wider driverless program.

Related, non-linked developments include:

• Thai police have arrested Singaporeans on human-trafficking charges; no case details were provided.
• Waymo plans to launch autonomous robotaxis in Singapore by 2028.

Neither report changes the immediate Punggol investigation.

Takeaway: Two current operators run three autonomous routes, while Waymo’s 2028 plan points to wider future competition.

What could the suspension change?

LTA may require changes to the AI, operating procedures, or both before testing resumes. Singapore wants to remain a global smart-mobility hub, but it must protect pedestrians before driverless services expand nationally. Any adjustment may become an international benchmark for urban autonomous operations.

Until investigators publish the report and LTA clears a resumption, the immediate question is narrower than the future of driverless transport: can Grab and WeRide show that their system protects a pedestrian when routine traffic rules and sudden human movement meet without warning?

Takeaway: Wider testing must rest on verified system performance, not company assurances alone.