Is Deep Reinforcement Learning the Next Advancement in Avatar Creation? Connecting the Dots From Artificial Intelligence to Motion Simulation

Image courtesy of DeepMotion

Image courtesy of DeepMotion

We’re not only creating avatars, we’re helping them learn too.

Diving into the world of avatar creation, Possible Reality is at the precipice of an emerging sector, developing proprietary technology that will allow users to create lifelike digital avatars using a single two-dimensional photo.

We’ve written at length about our INTARACT app -- a comprehensive AR ecosystem and marketplace for customizable digital avatars -- as well as our focus on B2B mixed reality solutions, but with the announcement of DeepMotion’s Neuron, which is currently in pre-release, it’s time to detail our emphasis on artificial intelligence, in particular facial recognition machine learning.

Up to this point, our AI efforts have been overshadowed by our larger mission to create lifelike digital avatars. However, artificial intelligence is an important aspect of avatar creation; it allows digital representations to learn applied behaviors rather than simply repeat pre-determined movements and facial expressions.

“Cloud-based AI is really going to change the future of growth in the VR/AR and robotics industries,” explains Possible Reality founder Ryan Isbell, referencing DeepMotion and their Deep Reinforcement Learning. “The use of machine learning allows characters to adapt to environments and objects both physically and behaviorally in real-time.”

While Possible Reality is perfecting facial recognition and mapping, so that avatars can move their mouths and adjust facial expressions, DeepMotion has been hard at work on its Neuron, which is backed by an algorithm that allows avatars to learn behaviors. They’re calling it “the first tool for completely procedural, physical character animation.”

Rather than use existing motion capture technology that houses a series of moves or combinations, Neuron allows avatars to learn a basic behavior set, much like their previous DeepDribble product teaches avatars how to seamlessly combine a number of skill sets associated with basketball. The result is a fluid movement from one skill to the next, although there is a learning curve for these avatars to master various movements.

Image courtesy of Venture Beat

Image courtesy of Venture Beat

Examining Neuron, along with DeepDribble, which was recently unveiled at SIGGRAPH 2018, it’s clear that DeepMotion understands the mechanics of movement and machine learning. They’ve also created a rather user-friendly editor, which allows users to adjust rigging points, which adds layers to further customization efforts, as these variables inevitably will affect an avatar’s ability to learn interactive behaviors.

However, with all of their advancements in movement, motion, and artificial intelligence, DeepMotion still has room to improve to customization options and capacity for their avatars, especially when trying to capture true digital likeness.

It’s this exact niche that has Isbell excited about the possibilities of combining DeepMotion’s Deep Reinforcement Learning with Possible Reality’s facial mapping capabilities.

“As avatar creators focusing on facial recognition machine learning, our multi-rigged avatar and agents creation tools can integrate with DeepMotion in a number of ways,” comments Isbell. “By combining our facial recognition technology with DeepMotion’s motion capture machine learning and AI capabilities, we can sell a large neural network of complete avatar creation and control.”

Image courtesy of DeepMotion

Image courtesy of DeepMotion

And therein lies major potential for advancing the capacity and functionality of AI-backed digital avatars.

While DeepMotion has developed a unique algorithm that allows avatars to learn motion, Possible Reality is filling in the gaps, focusing our artificial intelligence efforts on facial movement and tracking. And, by combining the two, avatars will be more realistic than ever before, complete with the capacity to learn complex body movements and detailed facial expressions.

Have an opinion on this blog? Interested in alpha testing our new AR and avatar products? We’d love to hear from you and share our exciting new products and comprehensive XR solutions.

For all blog inquiries, please contact

For all other inquiries, please contact