By: Andrew Ringer

Imagine what the world could be like if robots could replicate any motions that humans could make. Or if directors were able to tell a character how to act in an animated film the same way they could with a human actor.

These advancements are what new SFU computing science professor Jason Peng is working towards in his research. In particular, his doctoral dissertation focused on developing virtual agents that can replicate the motor capabilities of humans and animals.

To do so, he and his team use reinforcement learning techniques, a form of machine learning, to train simulated agents to perform complex motor skills by imitating demonstrations. This allows virtual characters to learn a large variety of skills, including acrobatics, martial arts, and even a little bit of breakdancing.

Despite the effectiveness of these techniques in training simulated agents, translating these capabilities to real robots remains challenging.

“Any motion that is fast moving and dynamic is still very difficult to reproduce in the real world on robots at this point,” says Peng.

While there could be many uses for robots in the real world eventually, Peng believes that his research is currently most useful in virtual domains, like computer animation.

“My hope is that instead of an animator having to hand animate or record all the motions for their applications, we can use simulations to automatically generate motions for the characters,” he says.

In recognition of his research, Peng will be receiving the SIGGRAPH Outstanding Doctoral Research Award this August at the annual SIGGRAPH conference, held in Vancouver this year.

“It’s nice validation to see that there is interest in this work after working in this field for so many years,” says Peng. 

“I hope that this helps draw some attention for simulation and machine learning techniques, and that this research may inspire some folks in industry to try these methods.”

Peng’s strong start to his research career can be partly attributed to his industry experience, including first getting into computer graphics through an internship at Capcom. To date, he has worked for NVIDIA, Sony, Google Brain, OpenAI, Adobe Research, Disney Research, Microsoft (343 Industries) and Capcom.

“Having connections to industry has been really useful to gain some insight into some of the more practical problems that people are facing in computer graphics,” says Peng.

After completing his PhD at UC Berkeley, Peng joined SFU as a professor due to the “strong visual computing reputation of SFU” and having family in Vancouver. Now, he is looking for MSc and PhD students to continue to advance his research area.

While Peng has achieved exciting results in simulated environments, he hopes to translate this research to the real world in the future in order to develop more physically capable robots. Applications of this could range from helping out with everyday chores, to construction, to performing tasks that are dangerous for humans.

Learn more about how prospective students can apply to work with Peng >>

Learn more about SIGGRAPH 2022 >>