ERGONOMICS IN DIGITAL ENVIRONMENTS:HUMAN SIMULATION CHALLENGES

HUMAN SIMULATION CHALLENGES

As human modeling becomes an integral part of the design process, the need for visual realism and analysis sophistication also increases. For better or worse, the visual appearance of human figures plays an important role in the acceptance of the technology and the perceived confidence of the results. Efforts in several areas focus on the increased realism of the human skin form. For perform- ance reasons, current commercial human models are ‘‘skinned’’ using polygonal segment represen- tations that are either completely static or pseudostatic. The figures are composed of individual segments, such as feet, lower and upper legs, pelvis, and torso. The segments are drawn as a collection of static polygons arranged to give the segment its anthropomorphic shape. Prudent selection of the shape at the ends of the segments allows joints to travel through the physiological range of motion without the creation of gaps. Pseudostatic skinning solutions ‘‘stitch’’ polygons between the nodes of adjacent segments in real time to avoid the skin breaking apart at the joints. These solutions can be made to look very realistic and are adequate for most ergonomic assessments and presentations. However, they do not model the natural tissue deformation that occurs at the joints throughout the range of motion. This is visually most noticeable at complex joints, such as the shoulder joint, or quantitatively at joints to which measurements are taken, such as the popliteal region of the knee. To better model these areas, a variety of methods have been described in the literature that deform the surface polygons according to parametric descriptions or underlying muscle deformation models (e.g., Scheepers et al. 1997). However, these methods have generally not been commercially adopted because they are computationally expensive and mostly unnecessary from the ergonomic analysis standpoint. Nevertheless, as the computer hardware capability increases and the availability of highly detailed whole-body-surface scans elevates the expected level of visual realism, these deformation techniques will become more prevalent.

Performance Models

Performance Factors

Performance models used in current commercial models are largely an amalgamation of models and data available in the ergonomics and human factors literature. As mentioned in the review of the performance models, the presentation of most of these research findings was originally not intended for integration into real-time simulation environments. The studies from which these data were de- rived also did not address some of the more contemporary ergonomic issues, such as the performance limitations of the elderly, cumulative trauma, shoulder injury, and movement modeling.

The aging population is elevating the need to have more specific performance models for this demographic. Questions of functional limitations resulting from decreased strength, reaction time, and joint range of motion all affect the design, both of products and workplaces. In the automotive design space, ingress / egress capability is an example of a task that may be influenced by these limitations. In the workplace, questions of strength and endurance need to be addressed. Cumulative trauma prediction presents a particular academic challenge because the etiology of the injury is largely unknown. Biomechanical factors clearly play a role but to date do not provide sufficient predictive power upon which to base a risk-assessment tool. At best, conditions associated with an increased likelihood of cumulative trauma can be flagged. Similarly, shoulder fatigue and injury prediction is not developed to the point where models incorporated into modeling software can accurately predict the injurious conditions. The significant social and economic cost of low-back injury has motivated considerable research in low-back modeling over the past 20 years. The research findings have re- sulted in sophisticated models and quantitative design guidelines and have allowed manufacturing organizations to reduce dramatically the incidence rates of low-back pain. Shoulder injury and cu- mulative trauma now need the same level of investment to mature the available data in these areas.

Variation Modeling

Even with the sophistication of the currently available biomechanical models, human model users are becoming increasingly interested in asking questions of these tools for which there are insufficient data. One such example is describing the expected population variability within the performance of a task. Each person will perform actions in a slightly different way, and these variations are not represented in models that describe an ‘‘average’’ response. However, human modeling simulation software is ideally suited to visualize this variability between people (i.e., data clouds). Future human performance and movement models may have this variability modeled so that it can be displayed in the human modeling environment.

Human Motion Control

One of the significant advantages contemporary human modeling tools provide in ergonomic assess- ments is the ability to assemble simulations of the workers performing their tasks. Simulations can add value for task-timing information, workcell layout optimization, training, and technical presen- tations. If we are confident of the motion realism, we can apply the posture-sensitive ergonomic assessment tools to help identify the situations with the greatest injury risk potential. Considerable effort has been spent searching for methods that accurately predict how humans move under different task and environmental conditions (Raschke et al. 1998). Dynamic simulation (Hodgkins and Pollard 1997; Popovic and Witkins 1999), statistical models (Faraway 1997), warping techniques (Bruderlin 1995; Witkins et al. 1995) and optimization (Chao and Rim 1973; Pandy and Zajac 1991) have all

been applied to this problem. However, many of the methods for simulating human motion and behavior are computationally intensive and do not lend themselves to real-time solution. While some methods show promise, no single method for modeling human motion has yet proven to be concise, flexible, and accurate. Modeling human movements accurately in constrained surroundings and when obstacles need to be avoided presents additional challenges.

Modeling Motion Data

Simulating human movements, whatever method is applied, requires a detailed understanding of how people really move. Much detailed research has been conducted in understanding lifting and arm movements, and the subject continues to be extensively studied (e.g., Chaffin et al. 2000). However, the wide variety of ways that humans can move and the flexibility we have to do the same task using different postural approaches create a challenge for trying to generalize these measurements to use in human modeling tools. It is one thing to measure the time sequence of joint angles involved in a typical lift task, but it is quite another to try to use these data to simulate accurately how people perform a lift under different conditions. Start and end conditions, the size, shape, or weight of the object being lifted, and obstacles that need to be avoided all influence the motion. Clearly a great deal of detailed data describing how we move under different circumstances is needed.

Multiple-figure Interactions

Humans do not necessarily work in isolation. Many tasks involve more than one human interacting with another. Two people carrying a single large load or manipulating the same object and one person slowing down or speeding up to avoid running into one another are just a few examples. All the motion-control challenges associated with modeling the movement of a single individual apply and are magnified when multiple individuals, each contributing differently to the task, are involved.

Interactive ‘‘Smart’’ Avatars Ultimately an accurate representation of humans needs to model not only how they move but how they think and make decisions about what movements to do and how they react to a given situation. Such ‘‘smart’’ humans would obviously aid in the gen- eration of a complex motion sequence involving several humans and have application to the devel- opment of workplace simulations. However, at this point in time, the development of intelligent human’s agents has been motivated by applications such as interactive training simulations (Badler et al. 1999), pedagogical agents (Johnson et al. 2000), intelligent characters in computer games (Funge et al. 1999), and conversational agents (Cassell and Vilhjalmsson 1999) and have not yet been applied to any great extent to workplace simulations.

2. CONCLUSION

Digital human modeling is being actively used in industries around the world to reduce the need for physical prototypes and create better and safer designs faster than was previously possible. Contem- porary human modeling software tools are actively assimilating a variety of previously disconnected human modeling knowledge, including population anthropometry descriptions and physical capability models. The large amount of ergonomic and anthropometric knowledge integrated into these solutions makes them efficient tools to answer a wide variety of human factors questions of designs. At the same time, the global nature of these tools is serving to consolidate and expose research findings from around the world and steering academic research direction and focusing the presentation of the results for model inclusion. While there are many areas that can be explored using the current offering of modeling solutions, many interesting challenges remain as we work to make virtual humans as lifelike as technology and our knowledge of humans allow.