New Computing Science Research Advances the Future of Intelligent Robotics

SFU Computing Science researcher is advancing how robots learn to move with the launch of MimicKit, a new open-source framework that makes it easier to train robots and simulated characters to imitate human and animal motion.

Developed by Assistant Professor Xue Bin (Jason) Peng, MimicKit brings together years of research on motion imitation and reinforcement learning into a single, accessible toolkit. The framework enables robots to learn complex motor skills, such as walking, dancing, acrobatics, and martial arts, by imitating motion demonstrations.

“Over the years, we have developed a number of motion imitation techniques that allow robots to learn highly dynamic and natural movements,” says Peng. “These techniques are already widely used across academia and industry. In fact, whenever you see a humanoid robot dancing or performing martial arts, there is a good chance it was trained using one of these methods.”

Lowering Barriers to Advanced Robotics Research

While motion imitation has proven to be highly effective, implementing these methods effectively has traditionally been difficult. Practitioners often need to make numerous subtle design decisions, and mistakes can lead to unstable or unrealistic motion. This complexity has limited who can use this technology and how quickly it can be applied in real-world settings.

MimicKit was designed to change that.

The framework provides high-quality, easy-to-use implementations of state-of-the-art motion imitation methods in one centralized, modular system. By packaging best practices and proven techniques into a single toolkit, MimicKit significantly lowers the barrier to entry for students, researchers, and engineers working in robotics, animation, and artificial intelligence.

Real-world Impact Across Industry and Society

The impact of MimicKit extends beyond academic research. Motion imitation is a key capability for real-world robots operating in human environments, from assistive and service robots to industrial automation and entertainment technologies.

In robotics, more natural and adaptive movement allows machines to work more safely alongside people, navigate complex spaces, and perform tasks that require balance and coordination. In computer graphics and digital media, these techniques support the creation of realistic animated characters for games, film, and virtual reality, reducing manual animation effort while increasing realism and interactivity. 

By operating in advanced simulation environments, MimicKit allows researchers to safely and efficiently test how robots learn to move before applying these methods in the real world. While the framework does not yet support deployment on physical robots, it is built with that future goal in mind. MimicKit also promotes collaboration and transparency, making it easier for researchers and industry partners to test ideas, compare results, and build on each other’s work.

Advancing Computing Science

As an open-source project, MimicKit reflects a broader commitment to advancing the computing science field through shared tools and community-driven innovation. It supports research in reinforcement learning, robotics, computer graphics, and embodied artificial intelligence, while also serving as a valuable educational resource for training the next generation of computing scientists.

“We hope MimicKit accelerates progress across the field,” says Peng. “By making motion imitation more accessible and reproducible, we are enabling new ideas, new applications, and new collaborations.”

This vision is already reflected in the works of SFU Computing PhD students Michael Xu and Ziyu Zhang, whose research leveraging MimicKit are featured at SIGGRAPH 2025 and SIGGRAPH Asia 2025.

With MimicKit, computing science researchers are helping shape a future where intelligent machines move more naturally, interact more safely with humans, and contribute meaningfully to society across industries.

This work was supported by NSERC, the National Research Council of Canada, and Sony Interactive Entertainment.

Authors:

• Xue Bin (Jason) Peng— Simon Fraser University
• Paper: MimicKit: A Reinforcement Learning Framework for Motion Imitation and Control.