Behavior registration and analysis for driver training on simulators
J.R. Kuipers
Green Dino Virtual Realities, Wageningen, The Netherlands
Registration of driver operations is essential for feedback systems, and for modelling the behavior of intelligent agents.
Virtual Observer
During Measuring Behavior '98, we presented "Virtual Observer" [1]: an
application that automatically registers user actions and other events in a
virtual environment and stores them in a log file. The stored data and corresponding
video information can then be analyzed with The
Observer [2]. One of the target markets for Virtual Observer is skills training
with simulators based upon virtual reality technology. Green Dino is now researching
new applications in cooperation with Wageningen University & Research Center
and Twente University.
Driver training
Driving schools all over the world are interested in training with interactive
simulators. Driving lessons become more and more expensive, and research shows
that the performance of a normal driving lesson is three times less effective
than a lesson on a simulator. In 1999, Green Dino began developing a driving
simulator called INTRASIM (Intelligent Training Simulator; Figures 1 & 2). A
major feature of this process involved using behavioral observation and registration
to provide feedback for the virtual instructor, and for the agent system that
calculates the behavior of the other traffic. Recording behavioral information
is necessary for both direct and indirect feedback systems.
Figure 1. INTRASIM from an external perspective.
Figure 2. INTRASIM from the driver's perspective.
Virtual Instructor
Most driving simulators offer a simple feedback system for instruction, based upon negative comments as a result of broken rules. We have built a new feedback system that registers behavior over a longer period in a log file. The software uses this information to give positive feedback. For example, the negative system will automatically give negative feedback on a fault action. The positive system relates this fault to the action history, and then decides if a response is necessary. The positive system can also compliment the driver when (s)he shows improvements. The virtual instructor uses the registered information for performance overviews and gives advice for future lessons. With The Observer, the registered data can be analysed quantitatively, together with the corresponding video information. This can be useful for researching the effect of the virtual instructor's comments on the driver's performance in the simulator.
Virtual Agents
In any driving lesson, other traffic is essential for driver training. The behaviors of the other vehicles must be interactive and respond to each other, and to the operator. Therefore, it is necessary to register the behaviors of all traffic members and send this information to the other members, and to the instruction system. Based upon this information, the agent system calculates new behavior for each traffic member. For behavioral calculations, we have developed a system with agents that have emotions. A new method will involve the use of live-recorded data; research shows that it is relatively simple to use live action data as input for traffic agents. Using The Observer, the live-recorded data will be translated into behavioral patterns, and a neural network will be trained in how to react to all kinds of traffic patterns. This neural network will be used by the agent to decide how to react to the behavior of other agents. Green Dino has entered into an exclusive partnership with the Parlevink group of Twente University to research the performance of this artificial intelligence system.
References
Paper presented at Measuring Behavior 2002 , 4th International Conference on Methods and Techniques in Behavioral Research, 27-30 August 2002, Amsterdam, The Netherlands