Technology
Waymo challenges Tesla’s self-driving approach in new blog post
Without naming its competitor, Waymo argues that camera-only systems and upgrading driver-assist software represent a "false summit" for full autonomy.
The short version
- Waymo's VP of Onboard Software, Srikanth Thirumalai, published a post outlining 10 AI lessons from the company's 200 million fully autonomous miles.
- The post critiques several core pillars of Tesla's autonomous strategy, including its reliance on camera-only hardware and its end-to-end neural network models.
- The critique comes as Waymo reports serving over 500,000 paid driverless rides weekly, while Tesla prepares to launch its dedicated Cybercabs.
Key facts
- Waymo's VP of Onboard Software and head of AI foundations, Srikanth Thirumalai, published a blog post outlining "10 AI Lessons from Driving 200+ Million Fully Autonomous Miles."[The Verge · Electrek]
- The blog post asserts that cameras alone are insufficient for full autonomy, contrasting with Tesla's camera-only "Tesla Vision" system.[The Verge · Electrek]
- Thirumalai writes that upgrading a consumer driver-assist system (Level 2) to full autonomy (Level 4) is a "false summit" and that true autonomy requires a purpose-built system.[Electrek]
- The post warns against using pure end-to-end neural networks, calling them "black boxes" that fail to build necessary trust.[Electrek]
- Waymo reports that it currently provides more than 500,000 paid, fully driverless rides each week across its U.S. markets, aiming to reach one million by the end of the year.[Electrek]
- Tesla recently confirmed that its unsupervised program has driven 380,000 unsupervised miles.[Electrek]
What remains uncertain
- It remains unproven whether Tesla's vision-only approach can successfully achieve generalized full autonomy to match or overtake Waymo's sensor-fusion system.[Electrek]