SHANGHAI — Automotive intelligence provider PATEO has integrated ByteDance’s Doubao language model and NVIDIA Nemotron into its latest in-vehicle computing unit, powered by the NVIDIA DRIVE AGX Thor platform.
The system utilizes an AI model suite designed to run complex inference tasks locally inside the vehicle without relying on an active cloud connection. The stack features ByteDance’s 35-billion-parameter Doubao Mixture-of-Experts (MoE) model alongside NVIDIA Nemotron 3.5 ASR for real-time speech recognition. To maintain low latency, the suite includes dedicated text-to-speech modules and a 2-billion-parameter model designed for semantic intent classification and request filtering.
The high memory bandwidth and accelerated compute of NVIDIA DRIVE AGX Thor enable the system to manage the expanding key-value (KV) cache generated during long, multi-turn conversational interactions. The integrated AI agent platform combines visual, voice, and vehicle sensor data to manage cabin routines, coordinate multi-zone climate settings, and evaluate driver fatigue during extended trips. Additionally, an integrated “Sentinel AI” mode monitors localized environmental conditions to issue driver alerts for road hazards, flooding, or fire risks.
Designed as a modular hardware add-on, the AI Box enables auto manufacturers to retroactively add generative AI processing capabilities to existing electrical architectures without replacing the vehicle’s central domain controller. The platform also establishes a framework for recurring automotive software services through on-demand feature deployments.
No Comment! Be the first one.