Tesla vs. Waymo: The Texas Robotaxi Fleet Gap

Waymo Driver (5th/6th Generation):

• Employs an integrated sensor suite including Lidar, Cameras, and Radar.
• Lidar: Provides a 3D picture of surroundings with 360-degree field of view and a range exceeding 300 meters, effective in all lighting conditions.
• Cameras: Offers a 360-degree vision system with high dynamic range and thermal stability, capable of identifying objects like pedestrians and stop signs from over 500 meters away. The Jaguar I-PACE fleet uses 29 cameras.
• Radar: Utilizes millimeter wave frequencies to measure object distance and speed, maintaining effectiveness in adverse weather conditions such as rain, fog, and snow.
• Onboard Computer: Acts as the 'brain,' combining server-grade CPUs and GPUs to process sensor data in real-time, identify objects, and plan safe routes using AI.
• Sensor Fusion: Software integrates diverse data inputs from all sensors (video, lidar point clouds, radar imagery) to build a comprehensive and coherent understanding of the environment.
• Simulation: The Waymo Driver has accumulated over 20 billion miles in simulation, alongside 200+ million miles of real-world driving experience, to train its AI and anticipate road user behavior.

Tesla Full Self-Driving (FSD):

• Vision-Based System: Primarily relies on an array of cameras for perception, a strategy known as "Tesla Vision," having previously removed radar and ultrasonic sensors from new vehicles.
• FSD Computer (Hardware 3 & 4): Utilizes custom-designed AI inference chips for processing.
• Hardware 3 (HW3): Manufactured by Samsung using a 14nm FinFET CMOS process, features over 6 billion transistors, and LPDDR4 RAM with 68 GB/s bandwidth. It includes two FSD Chips for redundancy, each with two systolic arrays capable of 36 trillion operations per second (TOPS), processing images at 2,300 frames per second.
• Hardware 4 (HW4): Introduced in January 2023, it is three to eight times more powerful than HW3, built on 10nm technology, and includes an upgraded camera suite and a new radar unit. FSD v13.2.1 (as of December 2024) is the first software version to leverage the native resolution of all HW4 cameras.
• Neural Networks: Employs 48 deep neural networks, requiring 70,000 GPU hours for a full build, to perform tasks like semantic segmentation, object detection, monocular depth estimation from raw camera images, and generating road layout and 3D objects from a bird's-eye view.
• Software Architecture: Features a centralized, model-based software architecture that enables rapid over-the-air updates and seamless integration of new hardware components.