Accession Number:



Drift Improvement with Reinforcement Training - Inertial Sensors - Year 1

Descriptive Note:

[Technical Report, Final Report]

Corporate Author:


Report Date:


Pagination or Media Count:



This report details the system, test environment, and results used to evaluate Reinforcement Learning RL algorithms for their ability to reduce the rate of drift in the positional error of an Inertial Navigation System INS without aiding from external sensors. A custom RL environment was created to train a RL algorithm to correct the raw inertial measurements from an INS in such a way that the position more closely matched the INS position after being corrected by a Global Navigation Satellite System GNSS. When GNSS aiding was removed, the RL system would continue to correct the inertial measurements as it was trained to do before GNSS aiding was removed. Multiple RL algorithms were used in the RL system, and their performance was evaluated on their ability to correct inertial measurements to allow for more accurate position solutions reduce positional error. The algorithms were also evaluated on their use of computer resources and ability to operate in real time. The data collections and evaluations described in this report show that a RL system can help reduce the positional error of an INS without aiding from external sensors such as GNSS. It also shows that some RL algorithms lend themselves better to this type of system than others. In the end, this research identified two RL algorithms that will continue to be used in further testing related to this work.

Subject Categories:

  • Navigation and Guidance

Distribution Statement:

[A, Approved For Public Release]