A new study into the effectiveness of an emotion-detecting VR headset has shown the device to be 93 per cent accurate.
The study took place at the London Science Museum over a six-week period. With over 900 participants, it was one of the one of the largest biometric datasets outside laboratory settings.
The goal of the project was to test the feasibility of detecting emotional state using a VR headset and platform designed by UK company, Emteq Labs.
A VR installation was set up on the second floor of the museum and allowed anybody to drop in and participate in the experiment, which put people through a range of different emotional stimuli, including exciting, soothing, happy and sad experiences.
Through machine learning, the data set was used to classify various emotional states, achieving 93% accuracy. Emteq Labs says the results demonstrate the potential of affective state detection to be utilised as a core component for healthcare and wellbeing applications.
Graeme Cox, CEO of Emteq Labs, says: “Our definition of emotional state is maybe a little bit different to the way that some companies and researchers work. Traditionally, there has been an approach of putting people into little boxes of emotion: happy, sad, disgusted, et cetera.
“But I’ve never really believed that our emotions work that way. In reality, we work across a plane of emotion which varies constantly.
“We build our emotional measurement model as two-dimensional Cartesian coordinates which have a dimension of arousal on the vertical; bored at the bottom and excited at the top. And then we measure valence, with positivity to negativity on the x coordinate.”
Usually, studies of this type are carried out within an academic institution, predominantly recruiting from the student population, who are generally of a similar age and learning background. In this study, Emteq Labs says it was able to involve a more diverse population of participants, tackling a long-standing issue of diversity in AI research data sets.
Emteq Labs designed three experimental scenarios within Virtual Reality (VR) based on an existing physical office room.
Each replica room was populated with audio-visual stimuli and props intended to evoke both positive and negative emotions, which Emteq’s algorithms aimed to detect using a range of sensors. These measured 32 data points from each individual, amounting to several million data points in total.
The data was recorded using a combination of movement sensors, heart rate sensors and electromyography sensors (detecting the electrical activity produced by skeletal muscles). The study also recorded emotional facial expressions from the VR mask itself.
Participants were asked to self-report how they experienced each situation which was compared against Emteq Labs’ predictions.
An interaction algorithm was implemented for the main study, to record and mark events throughout the participants’ experience with the immersive scenarios.
The VR experience had two modes of interaction which participants were separated into: active and passive. In the active VR scenario, stimuli were embedded within the 3D environments and the participant had complete control over their movement and duration of interaction with the stimuli surrounding them.
In the passive scenario, participants watched the experience of others as a pre-recorded video while seated on a chair, limiting their interaction with props and stimuli.
The data of both groups were then recorded continuously while exploring the three VR scenarios.
Emteq Labs is developing its VR headset as a potential treatment for mental health issues, with a focus on anxiety disorders. Its VR headset is able to place people in situations that trigger their anxiety, whilst monitoring their emotional response and adapting the scenario in real-time.
Emteq Labs are currently seeking NHS approval for the device and working towards clinical trials which are due to commence in 2021.
Cox says: “There are 300 million people worldwide suffering from clinically diagnosed anxiety and that’s aside from the extra billion who have undiagnosed anxiety.
“The NHS’s Long Term Plan is to see the number of people with anxiety disorders or depression who can access talking therapies increase by 380,000 per year to reach 1.9 million by 2023/24
“But predictions are showing that by the time they get there, the demand could be closer to the 3 million mark.
“So, the demand for therapy on the NHS vastly outstrips its ability to supply it. If we can deliver a model which enables more self-treatment and remote management therapy, then we can increase the amount of patients that the therapeutic process can handle.”
