When a Dropped Frame Becomes a Clinical Problem

When a Dropped Frame Becomes a Clinical Problem

A therapeutic virtual reality environment taught us valuable lessons about performance, comfort, and realism. Initially, we focused on creating an immersive virtual world, but soon realized that the most critical aspect was making the experience comfortable for patients undergoing de-addiction therapy. The key difference between game development and clinical VR lies in the user's purpose: while gamers seek entertainment, patients are there for therapy.

Even minor issues like frame drops can cause discomfort, making it essential to prioritize the user's experience.

One crucial realization was that building a 3D environment is not just about visual appearance. Traditional methods of visual editing may not be sufficient when the environment needs to change frequently. Instead, we chose to generate the entire environment through code. This approach has several advantages, including automatic adjustments when the structure changes and a clear understanding of design decisions. This allows developers to make informed choices without constantly questioning previous design decisions.

Motion sickness is a significant concern in clinical VR. The conflicting information between what the patient sees and what their inner ear perceives can lead to discomfort. Smooth camera movements, like continuous rotation, can cause discomfort for some users, particularly those new to VR. To alleviate this issue, we offered an alternative: snap turning. This method turns the view in fixed steps, which is less natural visually but can be much more comfortable for users.

In the outdoor environment, we aimed to create a natural experience by adding uneven terrain. However, this choice led to another problem: camera movements during walking that could cause discomfort. To address this, we controlled where the terrain appeared, keeping the path from the gate to the building level while making it more uneven toward the edges of the environment. To ensure accuracy, we measured the height variation along the actual walking paths.

When optimizing 3D assets for VR, it's crucial to remember that optimization should not compromise the visual representation of the object. We encountered a bush that looked wrong after simplification, with its foliage removed before the underlying structure. Instead of applying a uniform optimization to the entire model, we opted to treat different parts of the model separately and check the rendered results. This approach ensures that the model remains visually accurate while achieving optimization goals.

In some cases, the best solution may be to use a picture instead of a complex 3D model. For example, we opted to render ivy from several angles as transparent images rather than trying to create hundreds of thousands of triangles for a standalone VR scene. This alternative proved successful, as it maintained visual quality while avoiding the performance issues associated with complex 3D models.