Running AV Reasoning Models Within Onboard Compute and Latency Limits

NVIDIA introduces Alpamayo 2 Edge, an adaptive reasoning vision-language-action model designed for real-time autonomous vehicle operation on NVIDIA DRIVE AGX Thor. The model utilizes a 2B Nemotron reasoning backbone, diffusion action experts, and closed-loop reinforcement learning to generate chain-of-causation reasoning and future trajectories within strict onboard compute limits. Additionally, the company presents AlpaJudge, a human-calibrated evaluation and reward model used to assess reasoning traces and guide training workflows. Open article →