Autonomous Driving
Autonomous Driving: From PROMISE to REALITY - Part II

Let’s examine contradicting views:
█ A BELIEVER’s VIEW
Autonomous vehicles (AV) are no longer a bet… they are real:
• Waymo drove 170M miles in 5 cities
• Baidu completed 17M rides
• Aurora runs driverless heavy trucks on U.S. highways
• WeRide-Grab, Pony.ai-CDG are running autonomous shuttles in Singapore
And the safety data is stunning!
• Waymo’s dataset suggests 92% fewer injury claims vs. human-drivers
• A study in Traffic Injury Prevention in 2025 found 79% fewer injury-causing crashes
• Dr. Slotkin, a neurosurgeon, who analysed Waymo’s 100M mile data, wrote in the New York Times that deployment delays have “a body count” — that when results are this dramatic, the ethical calculus demands we treat them seriously!
Dr is right. Just in US, 39000 lives are lost to traffic crashes annually. Globally, the figure at 1.19 million deaths. If autonomous vehicles can cut that, even within limited ODD, we are morally obligated to move thoughtfully & urgently.
Now,
█ A CRITIC’s VIEW
We are systematically overstating these numbers.
All these statistics come from conditions that have been carefully curated - favourable weather, routes.
When someone reads “92% safer than human drivers,” they’re comparing performance in sunshine to sunshine.
KEY GAPs:
✘ Weather ODD Gap - Snow, dense fog, heavy rain
✘ AI Verification Gap - Black Box
⚠ Data Transparency Gap - Limited disclosure of remote operators assistance
⚠ Unknown-unsafe scenarios handling Gap
✘ Fleet-scale architecture safety thinking Gap - Recent incident of mass robotaxi malfunction in Wuhan could fall into this category
The AI system is the enabler of every AV are, at their core, black boxes. ISO 26262 covers HW & SW failures, ISO21448 & UL4600 cover performance insufficiency & safety case, ISO 8800 address AI safety. I have not seen any of the operators has publicly completed independent certification of ALL
📌 FINAL THOUGHT
Autonomous driving technology, within validated ODD, is truly safer than human drivers. The trajectory is real — LiDAR that costed $75,000 a decade ago now runs $1,000. Baidu’s RT6 platform costs $28,000. Huge improvement in the economics & safety. This is coming, and it should come, because thousands are dying on roads each year is not an acceptable baseline to defend.
A path forward requires:
• deployment at scale to uncover more & more edge-cases
• full transparency— in board rooms, in regulatory filings, & in conversations with the public — about safety to build trust
• addressing AI’s black-box verification limitations
• handing of unknown-unsafe conditions
The technology won’t fail us, temptation to overstate it might.
In Part III & beyond, I’ll cover COST, TECH, REGULATION…
- April, 2026
