Answers grounded in memory
GBR reads knowledge stored on the device without calling an external server. Sources accompany its answers.
GBR SYSTEM DEMONSTRATION
Meet GBR SYSTEM, built on its own architecture, on a real device. It answers from its own memory and remembers new knowledge taught through conversation.
Filmed on one NVIDIA Jetson Orin Nano Super with the network disconnected.
In this demo, questions and motion commands are typed on a keyboard. Active tracking and speech recognition are in development.
GBR reads knowledge stored on the device without calling an external server. Sources accompany its answers.
A user teaches a new fact and confirms that GBR should remember it. GBR uses that memory in the next answer without restarting or retraining.
A retracted memory is no longer used in answers. The record of its retraction remains.
It understands the words it was taught and controls the camera and motor. Answering questions and moving hardware happen on the same device.
The latter part of the video presents the Sudoku results of the GBR reasoning engine. Explore the published results and materials on the research page and GitHub.
GBR SYSTEM is made of GBR DBMS, which stores and finds memories, and GBR AI, which handles everything from perception to generation. Memory and reasoning connect within one system.
Let's explore applications for robots and embedded devices together. Tell us about the knowledge and device requirements you need.