Starship's Ahti Heinla: 8 million deliveries, zero stolen robots, and why sidewalk bots don't need lidar

Summary

Ahti Heinla, the co-founder of Starship Technologies and one of the original engineers behind Skype, is running what he describes as a fully commercial sidewalk delivery robot operation — not a pilot. Starship has completed 8 million deliveries across 60 cities and 60 college campuses in six countries (five in Europe plus the US), partnering with major delivery apps and tier-one retailers. In the markets where it has deepest penetration, robotic delivery has become the default — residents don't use human couriers.

Zero stolen robots

Vandalism, the obvious skeptic's concern, simply hasn't materialized. Across all 8 million deliveries, no Starship robot has ever been stolen from the street. Heinla attributes this partly to design — the robots carry 10 cameras, maintain a constant internet connection, and trigger a loud siren when tampered with — but also to something less engineered: people treat the robots warmly. Children feed them bananas. The dynamic is closer to a UPS truck (which nobody punctures) than to a dockless scooter (which people do abuse).

Why no lidar

Starship's robots do not use lidar, but Heinla is clear this is a functional mismatch, not ideology. Lidar is optimized for long-range detection at speed — useful for a car moving at highway velocity, less useful for a robot crawling a sidewalk. The more fundamental problem is vertical field of view: lidar's narrow scanning angle is poorly suited to the close-range, wide-angle vision a sidewalk robot needs to see curbs, feet, and obstacles immediately in front of it. If a lidar with the right specs becomes available, Heinla says Starship will use it.

Hybrid autonomy, not end-to-end AI

Built over 10 years, Starship's system combines C++, neural networks, and remote human assistance — the latter becoming less frequent as AI improves but still present for unusual situations. Heinla is skeptical of competitors claiming pure end-to-end neural network operation, not because he doubts it's technically possible, but because it is very hard to prove it is safe. Starship's robots cross roads 100,000 times a day. At that volume, even a 1% failure rate is unacceptable, and a hybrid architecture makes the safety case far easier to demonstrate to regulators. The broader AI wave — transformers, improved vision models — has helped reduce the need for human-in-the-loop intervention, but Heinla says Starship was already at commercial quality before those advances arrived.

Competitive moat

The closest competitor has done roughly 200,000 to 300,000 deliveries — an order of magnitude behind Starship's 8 million. Heinla's point is structural: whatever is hard for Starship is equally hard for anyone entering later, and 10 years of operational data across six countries is not easily replicated. The demand side is not the constraint. A major delivery app founder told Heinla directly that if robots cost less than human delivery and actually work, they'll use them for billions of deliveries. Starship's job is to keep proving the unit economics.

The Skype parallel

Heinla co-wrote Skype with roughly 15 engineers over nine months — a product that went viral from day one. The lesson he carries forward is that if a product fits exactly what people need and has no major downside, adoption is not the problem. Robots are harder to build than VoIP software, but that cuts both ways: the barrier is also higher for everyone else. Skype shuts down on May 5th; Heinla hasn't used it in years and takes the closure in stride. He built it more than 20 years ago.

Estonia's startup culture

Heinla turned 19 the year Estonia regained independence from the Soviet Union. His explanation for why the country produces disproportionate tech talent is simple: a newly free country with no entrenched corporate establishment gives founders no reason to believe anything is impossible. Not knowing what's impossible, he argues, is a durable competitive advantage.