Disney Research is a privately owned entity, headquartered in the US, and operates under Disney Experiences. Founded in 2008, employing approximately 100 individuals, the company offers research services as a corporate network of scientific and technological research laboratories supporting the multi-faceted business segments of The Walt Disney Company. It operates primary facilities in Los Angeles, United States, and Zurich, Switzerland (including specialized arms such as DisneyResearch|Studios), alongside an academic partnership with ETH Zurich. Functioning as an internal innovation hub rather than a consumer-facing media vendor, it tasks international researchers with developing early-stage scientific concepts to enhance corporate assets. Its core areas of investigation encompass computer vision, robotics, machine learning, human-computer interaction, visual computing, and immersive technologies, generating proprietary hardware and software models—such as adaptive character controls, robotic motion diffusion frameworks, and omnidirectional locomotion surfaces—that are integrated across Walt Disney Studios productions and Disney Experiences theme park attractions.
On July 2, 2026, Disney Research revealed its work on bringing the animated character Olaf into the physical world, relying on reinforcement learning guided by animation references for control. This initiative addresses the challenges posed by animated characters' non-physical movements and proportions, which differ from typical walking robots. The day prior, on July 1, the research division focused on "Neural Render Proxies for Interactive and Differentiable Lighting" to streamline the CG animation production pipeline. This aims to reduce the significant iteration times caused by slow offline renderers and complex shader evaluations required for even minor lighting adjustments. Previously, on May 13, Disney Research introduced "CoCo-InEKF," a differentiable Invariant Extended Kalman Filter. This system utilizes a neural module to predict contact velocity covariances, enabling robust state estimation in dynamic, contact-rich scenarios and achieving improved performance over classical and other learning-based approaches. On May 12, the team shared a method to improve robot movement accuracy, proposing a bilevel optimization framework that jointly adapts human kinematic reference motions to a robot's morphology while training a tracking policy via reinforcement learning, aiming to prevent physical inconsistencies such as foot sliding or dynamically infeasible motions. This followed an earlier demonstration on May 10 of "ReActor," a new motion-learning system designed to help robots move more like humans by transferring human motion to robots with diverse body shapes and movement capabilities.
Subscribe for full access to Disney Research's products in full detail
Subscribe for full access to Disney Research's revenue in full detail
Subscribe for full access to Disney Research's products in full detail 
