Analysts Find Waymo Outperforms Tesla FSD
In a head-to-head test of autonomous vehicles in Austin, Jefferies analysts found Waymo's service superior to Tesla's FSD in both ETA accuracy and overall efficacy. While Tesla was the cheaper option, the analysis suggests hybrid fleets with human drivers may dominate near-term adoption to manage costs and wait times for fully autonomous rides.
The Jefferies analysis involved 19 Uber rides, 14 of which were fulfilled by Waymo, and 15 rides in Tesla robotaxis within their designated operational zones in Austin. This direct comparison revealed that while Tesla offered a significant 60% discount compared to UberX, it underperformed in key service metrics. Tesla's vehicles exhibited significantly longer wait times and took suboptimal routes, leading to longer trips. The analysis also highlighted that both services struggled with precise pickup and drop-off locations, a persistent challenge in autonomous ride-hailing. Despite Waymo's operational expansion to a 90-square-mile area in Austin, wait times were still longer than a typical human-driven Uber. The performance differences are rooted in fundamentally different hardware and AI philosophies. Waymo's vehicles utilize a comprehensive sensor suite that includes 29 cameras, LiDAR, and radar, and rely on extensively pre-mapped, geofenced areas for operation. In contrast, Tesla employs a vision-only system with 8 cameras, foregoing LiDAR and radar entirely, aiming for a more generalizable solution that isn't dependent on high-definition maps. On the software side, Tesla's FSD Version 12 marked a radical shift, replacing over 300,000 lines of C++ code with an "end-to-end" neural network. This system learns driving behavior by analyzing vast amounts of video data from its fleet, moving from explicit rules to imitating human driving patterns. This approach is designed to interpret scenes and generate control outputs directly from pixel data. Waymo employs a "Foundation Model," a large-scale AI trained on extensive sensor data, to simulate scenarios and inform its driving decisions. This massive model then "teaches" smaller, more efficient models that run in the actual vehicles, allowing for real-time perception and prediction without constant cloud reliance. This architecture is designed for robust safety validation and to handle a wide array of complex driving environments. For aspiring engineers in this field, the technical stacks of these companies highlight key skills. Proficiency in C++ and Python is crucial, alongside experience with the Robot Operating System (ROS). For embedded systems roles, expertise in real-time operating systems (RTOS), sensor fusion (combining data from sources like LiDAR, radar, and cameras), and developing control algorithms for hardware is in high demand. The evolution of AI in this sector is trending towards models that can handle long-tail edge cases and reason more effectively in novel situations. Experience with machine learning frameworks like TensorFlow and PyTorch, deep learning, and computer vision are foundational for developing the neural networks and perception systems at the core of both Waymo's and Tesla's platforms.
