Waymo CEO explains why Tesla's camera-only self-driving falls short

4 hours ago

Dmitri Dolgov argues that camera-only systems cannot achieve the superhuman performance required for full autonomy, as weak sensing leads to a safety ceiling.
Waymo uses a combination of cameras, lidar, and radar because each sensor compensates for the others' weaknesses, particularly in darkness, glare, fog, and physical obstructions.
Cameras can match human driving performance but plateau before reaching the reliability needed for driverless operation, unlike multi-sensor fusion.
The 'nines' problem illustrates that each additional nine of reliability requires roughly ten times more effort, making camera-only systems hit a plateau early.
Tesla's camera-only robotaxi service has a crash rate three times worse than human drivers, while Waymo's multi-sensor system shows 94% fewer serious-injury crashes.
Cost of lidar is decreasing rapidly with each hardware generation, undermining the argument that camera-only is more pragmatic.

Current AI agents achieve only low completion rates on realistic computer tasks (e.g., 20.6% on OSWorld-V2, 26.2% on Agents' Last Exam), far below user expectations set by LLM chat performance.
Benchmarks like OSWorld and WebArena gave inflated results because they used static, clean environments with simple action spaces, unlike messy real-world interfaces.
The core flaw is that agents often bypass the UI entirely (e.g., using API calls instead of clicking), leading to inefficiency, side effects, and unreliable performance on tasks that require direct GUI interaction.

AI agents are projected to generate $450 billion in economic value by 2028, but deployment remains in single digits across business functions.
Coding agents are an exception, successfully shipping end-to-end work, while most other agents struggle with reliability.
Four key factors explain the gap between promised and delivered value: variance, benchmark quality, agent-specific errors, and alignment.

Tesla's latest iOS app update contains code for a new driver identity verification feature using the cabin camera to allow Full Self-Driving activation.
If the camera fails to match the driver to an authorized profile, FSD will be blocked with an error message shown in the app.
The code was discovered in version 4.58.5 via decompilation and includes strings like 'fsdIdentityCheckFailedTitle' describing the flow.

Waymo's driverless vehicles had 68% fewer crashes per mile compared to human drivers in four U.S. cities.
The Insurance Institute for Highway Safety (IIHS) study highlights the need for better federal data monitoring for automated vehicles.
Waymo operates Level 4 automation, which can drive without human supervision under certain conditions, unlike Level 2 systems that require constant driver attention.

The author trained a Joint-Embedding Predictive Architecture (JEPA) model, named LeMario, on Super Mario Bros to learn world dynamics from pixels and actions.
The model included a vision encoder, action encoder, and causal predictor with Adaptive LayerNorm Zero (AdaLN-Zero) for action conditioning, and used SIGReg to prevent representation collapse.
LeMario outperformed baselines in short-horizon prediction on held-out episodes, showing it learned action-conditioned dynamics.

Current AI agents achieve only low completion rates on realistic computer tasks (e.g., 20.6% on OSWorld-V2, 26.2% on Agents' Last Exam), far below user expectations set by LLM chat performance.
Benchmarks like OSWorld and WebArena gave inflated results because they used static, clean environments with simple action spaces, unlike messy real-world interfaces.
The core flaw is that agents often bypass the UI entirely (e.g., using API calls instead of clicking), leading to inefficiency, side effects, and unreliable performance on tasks that require direct GUI interaction.

AI agents are projected to generate $450 billion in economic value by 2028, but deployment remains in single digits across business functions.
Coding agents are an exception, successfully shipping end-to-end work, while most other agents struggle with reliability.
Four key factors explain the gap between promised and delivered value: variance, benchmark quality, agent-specific errors, and alignment.

Tesla's latest iOS app update contains code for a new driver identity verification feature using the cabin camera to allow Full Self-Driving activation.
If the camera fails to match the driver to an authorized profile, FSD will be blocked with an error message shown in the app.
The code was discovered in version 4.58.5 via decompilation and includes strings like 'fsdIdentityCheckFailedTitle' describing the flow.

Waymo's driverless vehicles had 68% fewer crashes per mile compared to human drivers in four U.S. cities.
The Insurance Institute for Highway Safety (IIHS) study highlights the need for better federal data monitoring for automated vehicles.
Waymo operates Level 4 automation, which can drive without human supervision under certain conditions, unlike Level 2 systems that require constant driver attention.

The author trained a Joint-Embedding Predictive Architecture (JEPA) model, named LeMario, on Super Mario Bros to learn world dynamics from pixels and actions.
The model included a vision encoder, action encoder, and causal predictor with Adaptive LayerNorm Zero (AdaLN-Zero) for action conditioning, and used SIGReg to prevent representation collapse.
LeMario outperformed baselines in short-horizon prediction on held-out episodes, showing it learned action-conditioned dynamics.

Hasty Briefsbeta