NVIDIA outlines a full-stack architecture for building agentic, multimodal AI assistants for vehicle cabins. The post covers hardware options (DRIVE AGX Orin, Thor, AI Box, MediaTek Dimensity AX pairing), a hybrid edge-cloud inference model, and the software pipeline components needed — ASR, LLM/VLM inference via TensorRT Edge-LLM, orchestration via NeMo Agent Toolkit, and TTS. It also describes the development workflow from cloud-based AI factory (NeMo, NIMs) through quantization and pruning to on-device deployment, with guidance on getting started and a list of ecosystem partners.
Table of contents
AI box: A dedicated platform for in-vehicle AI LLM accelerationDRIVE AGX Thor: Multi-domain AI computerCentral car computer with DRIVE AGX and MediaTek Dimensity AXHybrid architecture: AI inference from cloud to edgeBuilding a hybrid in-vehicle agentic AI pipelineFrom AI factory to in-vehicle deploymentDeploying on edgeGetting StartedSort: