/Patent Tracker
NvidiaPatent applicationCars & transportInterest 6/10

Nvidia patents hybrid system mixing rules and AI for self-driving

Nvidia describes a control system for autonomous vehicles that combines traditional rule-based logic with a neural network planner. Both parts share a common data-processing core, allowing the system to use established safety rules while learning from experience. The filing outlines a two-step training method to integrate these different control styles into one hybrid stack.

Why it mattersThis approach could help self-driving cars balance the reliability of known safety rules with the adaptability of machine learning, potentially making automated driving more robust in complex situations.

In the filing’s own words

In various examples, a hybrid control stack may be implemented with classical and E2E signal paths that share a common feature extractor or backbone network. In a first training step, a robust feature extractor or backbone network may be pre-trained by co-training the network sequentially using the different types of training data for each detection head and the neural planner—without the classical refinement algorithms. In a second training step, the neural planner may be fine-tuned with the common feature extractor or backbone network frozen. As such, the present techniques may be used to transition from a decomposed control stack to an end-to-end control stack, adapt an existing decomposed control stack into a hybrid control stack, train the hybrid control stack, and/or control navigation of autonomous vehicles, semi-autonomous vehicles, robots, and/or other object or machine types.

What exactly they claim (claim 1)

1. One or more processors comprising processing circuitry to: process a representation of sensor data corresponding to an environment of an ego-machine using one or more encoder networks to generate a scene embedding corresponding to the environment; generate, based at least on processing the scene embedding using a modular signal path comprising a rule- or heuristic-based planner, a first candidate action; generate, based at least on processing the scene embedding using an end-to-end signal path comprising a neural planner, a second candidate action; and control one or more operations of the ego-machine based on at least one of the first candidate action or the second candidate action.

Facts

Company
Nvidia (NVIDIA CORPORATION)
Official title
HYBRID DRIVE STACK WITH END-TO-END LEARNING FOR AUTONOMOUS AND SEMI-AUTONOMOUS MACHINES AND APPLICATIONS
Publication number
US20260274297A1
Application number
19078542
Filed
Mar 13, 2025
Published
Sep 17, 2026
Inventors
8
Claims
20

Classified under

B60W 60/0011 Operations & transportG06N 3/0455 Physics & computingG06V 10/82 Physics & computing
Read the full filing (official PDF) ↗ Open in the Patent Tracker

More from Nvidia in Cars & transport

Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.