In recent years, driver education and training has been revolutionized by the low cost and widespread availability of virtual learning systems. From simple PC- based driving tutorial programs to large scale advanced driving simulators, new av- enues for both education and research have emerged. Boyle and Lee [74] reported that there were 124 papers published between 1965 and 1999 (3.7 papers per year) with the words “driving simulator” in the title, abstract, or topic according to the ISI bibliometric database. In comparison, there were 572 papers published between 2000 and 2009 (63.5 papers per year). Driving simulators can now be found in a variety of configurations boasting different features and advantages. Subsequently, many researchers have put a lot of effort into validating the results they obtain from particular simulator systems.
No matter how complicated and advanced a driving simulator is, it is impor- tant to establish the extendability of results and conclusions to real life. Ideally, the best results would come from a simulator capable of replicating reality to the fur- thest extent. This however, inherently adds costs and complexity possibly beyond the capabilities of many research projects. Many researchers have shown that there is plenty of relevant and applicable information that can be obtained from lower level simulation systems. Desktop driving simulators offer the best in terms of low cost and mobility, however, as one of the most basic designs, they sacrifice a good amount of realism and limit the application of results. Allen et al. [75] compared a single monitor desktop simulator with a narrow field of view display, a triple monitor desk- top simulator with a wide field of view display, and a vehicle cab simulator with wide field of view projector display for novice driver training. The authors note that the desktop systems were ideal for implementation in high schools, however, the single monitor simulator was associated with the poorest performance most likely because of the narrow field of view. Another low-cost mobile driving simulator was developed in [76] for novice driver safety training. The inclusion of a racing seat with seat belt, as shown in Figure 2.3, provides a increased sense of realism over the desktop configurations and shows promising results for improving driver performance.
Increased costs and some sacrifice in mobility opens the world of driving sim- ulators featuring vehicle cabins for drivers to sit in and multiple projector screens or large monitors for display as shown in Figure 2.4. Fixed-based cab simulators offer a convenient balance in cost, complexity, and an immersive environment for added realism and as such have become very popular in implementation. Kaptein et al. [77] investigated the validity of a mid-level fixed-based driving simulator in assessing driving behavior. Their results showed relative validity for behavioral variables such as speed choice and lane-keeping performance, indicating that such results while valu-
Figure 2.3: The Clemson Automotive Training System (CATS) is an example of a low-cost mobile driving simulator
able should be interpreted carefully. Absolute validity was found for strategic route choice decisions as drivers showed the same choice behavior in real traffic as they did in the simulator. In [78], a fixed-based simulator was used to help novice drivers acquire higher-order perceptual and cognitive skills for safe driving. Drivers who re- ceived simulator based road hazard handling training showed earlier hazard detection, improved handling performance, and lower overall mental workload as compared to untrained drivers in a simulated test environment. Underwood et al. [79] conducted a driving simulator validation study in the context of hazard perception. Participants were evaluated while driving in real traffic, while watching film clips recorded from a real vehicle in traffic, and while driving in a simulated traffic environment. The results validated the fixed-based simulator as increased scanning and earlier eye fixations on
Figure 2.4: Example of a fixed-base vehicle cabin driving simulator
hazardous objects was detected for more experienced drivers in all three scenarios. Additional studies involving fixed-based simulators are mentioned in [80, 81, 82, 83]. Motion-base driving simulators represent an effort to create a highly realistic driving experience and come with an increase in complexity and cost. The motion of the simulator base can range from a simple single degree like that mentioned in [84] all the way up to more complicated 6 DOF and even 9 DOF platforms. Some of the most advanced simulators in the world include the National Advanced Driving Simulator (NADS) developed at the University of Iowa [85], The University of Leeds Driving Simulator (UoLDS) [86], and Toyota’s Driving Simulator at the Higashifuji Technical Centre in Japan [87]. These systems supply some of the highest fidelity driving simulations ever built and they have been used for numerous validation, behavioral, vehicle development, training, and other automotive related studies [86, 88, 89].
Along with validation studies, the numerous implementations of driving sim- ulators has also prompted various publications focused on evaluating simulator tech- niques and providing recommendations for improving results. In [84], Norfleet et al. compares a desktop simulator, a fixed-base in-vehicle cab simulator, and a mid-level motion based cab simulator with haptic steering feedback. The hardware, software, and general features of each simulator are rated and evaluated within the context of Research, Education, and Entertainment along with recommendations for implemen- tation in particular applications. Green et al. [90] examines the problems commonly found in fixed-base simulators which lead to inapplicable results. Green offers recom- mendations and solutions to problems such as unlimited driving boundaries, driving too fast, and lack of handling and/or road imperfections. Another important condi- tion for validity of driving simulator based experiments is adaptation. Sahami and Sayed, in [91], investigated adaptation times in a driving simulator using a power curve to mathematically model learning patterns of subjects. They found that adap- tation time and learning rate was task-independent and not significantly different for males versus females. They also recommend that practice scenarios should include the use of all control inputs with repetition, if possible, to allow iteration and adjustment of strategies. Changing conditions and scenarios are better than static conditions to ensure and also detect that adaptation has occurred.