From Segment Anything (Virtual AI) to Autonomous Trucks (Physical AI)

Tete Xiao, VP of Engineering and AI, Bot Auto joined Grayson Brulte on The Road to Autonomy to discuss the fundamental shift from virtual AI to the physical AI required for commercial autonomous trucking.

Tete co-authored Segment Anything, the landmark paper that ushered in the era of specific models to an era of foundation models that generalize across large segments of data. This approach which he is implementing at Bot Auto, enables the company to move beyond the limitations of previous technology, treating autonomous trucking as a compute-driven challenge where the system learns to navigate the complex physics of driving a truck.

To ensure safety, Bot Auto is utilizing a top-down redundancy architecture that mirrors aviation’s triple autopilot systems. Including dual onboard computers and independent software stacks running parallel algorithms with deliberately different logic to prevent a single failure from propagating through the system.

This spring, Bot Auto is planning to launch fully autonomous commercial operations with Ryan Transportation on the Houston to Dallas corridor. No safety driver. No safety observer. No human in the cab.

Episode Chapters

• 00:00 AUTNMY AI
• 00:25 Segment Anything
• 05:04 Virtual AI to Physical AI
• 09:08 Redundancy and Aviation-Inspired Architecture
• 13:40 Hardware and Software
• 17:00 Launching Fully Autonomous Operations
• 20:00 Foundation Models and Reinforcement Learning
• 27:52 Compute Infrastructure
• 35:22 Staying Ahead
• 42:30 Building a Virtual Driver
• 47:06 AGI
• 48:36 Transportation Company
• 53:59 Future of Bot Auto