Improving Intelligent Tutoring of Ship Handling
Abstract
Improving Intelligent Tutoring of Ship Handling. The objective of the proposed research is to increase the across-the-board effectiveness of COVE ITS at teaching ship handling fundamentals to match or surpass that of an average instructor. Previous tests of COVE ITS have shown it to be equally effective overall, but in certain areas to be less effective than average structors. Some shortfalls can be remedied by improving the match of COVE ITS’s tutoring to a given error’s most common causes. For other problematic cases, causes of the ship handling error vary over a wide enough range that COVE ITS’s tutoring needs to be more finely adapted to individual learners’ probable reasons for erring. This project will yield an improved COVE ITS that will be tested at SWOS to validate its effectiveness across the range of ship handling fundamentals. Prior research on intelligent tutoring in the dynamic domain of ship handling resulted in a prototype COVE ITS system. The system functions by monitoring people as they use their current ship handling knowledge to conn a simulated ship. It coaches them when it detects the ship is outside acceptable bounds of position, heading, speed, engine or rudder settings, etc. Using a synthetic voice, COVE ITS gives the person information or suggestions keyed to the most common reasons that learners put the ship in the unacceptable situation detected. This system was tested to compare its effectiveness with that of SWOS’s uniformed instructors at teaching junior officers to (i) get a ship under way from a pier, (ii) moor a ship to a pier under mild conditions of wind and current, and (iii) carry out underway replenishment. Analysis of these controlled experiments revealed that the COVE ITS prototype was as effective overall as the average instructor at teaching fundamentals in the simulator. These tests also showed that the software was less effective than live instructors in certain important areas, including helping learners avoid ‘pier shyness’, which can lead to stalling out while approaching a pier to moor, and helping officers learn to identify, position their ship in, and keep it within Waiting Station, a compact invisible ‘moving box’ located a fixed distance behind and offset from their supply ship. Henceforth we refer to these areas – the ship handling errors and their causes along with the prototype’s tutoring of them – as shortfalls of COVE ITS in achieving parity with average uniformed instructors at SWOS in coaching all ship handling fundamentals. We analyzed situations in which officers that COVE ITS tutored learned skills less well than those whom live instructors taught (i.e., shortfalls). Findings suggest that COVE ITS’s lesser tutoring effectiveness in these cases resulted from an inadequate match of the coaching it provided to the actual reasons a significant number of learners made the error. Two sorts of problem case appear to exist. (1) In one class, enhancing COVE ITS-delivered information or suggestions to better match the most common causes of the detected error seems likely to rectify the system’s shortfall in effectiveness. (2) The other sort of problem case is more complex. In these cases, learners’ reasons for erring vary widely and appear to be different enough that helping a majority of poorly served learners probably requires COVE ITS to specialize information and suggestions to several distinct causes of the error. This entails a requirement that the system detect students’ reasons for erring in these cases. Our hypothesis is that enhancing the prototype system in the two ways just described will increase its effectiveness with the test population to equal that of an average instructor across the board of ship handling fundamentals
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Nov 23, 2016
- Source ID
- N000141612414
Entities
People
- Stanley Peters
Organizations
- Office of Naval Research
- Stanford University
- United States Navy