Advanced algorithm capable of evaluating trainee progress, improving training quality and providing prognostics estimation for the trainee success - INT-007

Preferred Disciplines: Computer Science, Machine/Deep learning (PhD)
Project length: 12-18 months (3 units)
Approx. start date: February 2019
Location: Tel Aviv, Israel
No. of Positions: 1
Preferences: None
Company: Elbit

About Company:

Elbit Systems Ltd. is an Israel-based international defense electronics company engaged in a wide range of programs throughout the world. They are a leading provider of simulators in Israel and one of the most innovative simulator companies in the world.

In 2016, Elbit Systems had approximately 15,000 employees, the majority of whom are engaged in engineering, mechanics, research and development, and other computing and technical areas. Elbit Systems' shares are traded on the Tel Aviv Stock Exchange and NASDAQ

Summary of Project:

Training pilots is a very expensive task. Therefore, using simulators, rather than real flights, is very common.

High-end trainers provide a complete imitation of the cockpit and a close-to–real flight experience. We want to add real time trainee performance evaluation to improve debriefing, training quality and trainee evaluation as follows:

  • Analysis of progress patterns and detection of difficulty points allows better debriefing and capability to adjust training in order to focus on improving the trainee performance at weak spots.
  • Analysis of the performance of many trainees to learn performance / progress patterns shall enable better prediction of future achievements of a cadet based on his past performance.
  • Analysis of trainee’s physiological parameters Vs performance shall enable prediction of trainee performance under stress conditions.

Research Objectives/Sub-Objectives:

  • During training, a huge amount of data is collected. Including platform related data (position, orientation, velocities, accelerations etc.) as well as trainee data (physiological parameters, response times etc.). Elbit would like to leverage machine learning / deep learning techniques in order to analyze the data.
  • Use reinforcement learning methods to find properties in the data we collect from the trainers. (Currently reinforcement learning tools do not give sufficient emphasis to solving problems with constraints, and are usually used to find optimal solutions to problems without limitations). As such, an objective would be to develop a reinforcement learning-based technique that emphasize multiple constraints, specifically: training according to flight doctrine (e.g. how to land a helicopter in a challenging environment).
  • Quantify the pilot’s performance, in order to be able to compare it to the doctrine, which is considered the optimal operating profile. This will allow debrief (if required – in real time) based on the gap between the trainee’s performance and the expected performance.
  • Perform a long-term analysis of training data of cadets, starting at early stages of their practice and until they become experienced pilots/operators. Use the results of the analysis to predict - regarding a specific pilot and based on his performance in the early stages of the practice – if he will complete the training successfully.
  • Learn about correlations between physiological parameters of the trainee to the level of their performance. This will allow to predict performance at various stress levels.

Methodology:

  • To be discussed

Expertise and Skills Needed:

  • To be discussed

For more info or to apply to this applied research position, please

  1. Check your eligibility and find more information about open projects.
  2. Interested students need to get the approval from their supervisor and send their CV along with a link to their supervisor’s university webpage by applying through the webform.
Program: