Olaf: Bringing an Animated Character to Life in the Physical World

TL;DR

The paper presents Olaf, a costumed robotic character realized with an asymmetric 6-DoF leg design and hidden actuators, controlled via reinforcement learning guided by animation references to achieve believability. To prevent overheating and reduce footstep noise, the authors introduce a thermal-aware policy using actuator temperature as an input and a foot-impact reduction reward, implemented with a control-barrier-function model for thermal and joint limits. A two-layer system separates the RL backbone from show functions, and a path-frame approach ensures animation consistency. Real-world experiments demonstrate tracking accuracy around 4 degrees and notable reduction in footstep noise, establishing Olaf as a believable animated character in the physical world.

Abstract

Animated characters often move in non-physical ways and have proportions that are far from a typical walking robot. This provides an ideal platform for innovation in both mechanical design and stylized motion control. In this paper, we bring Olaf to life in the physical world, relying on reinforcement learning guided by animation references for control. To create the illusion of Olaf's feet moving along his body, we hide two asymmetric legs under a soft foam skirt. To fit actuators inside the character, we use spherical and planar linkages in the arms, mouth, and eyes. Because the walk cycle results in harsh contact sounds, we introduce additional rewards that noticeably reduce impact noise. The large head, driven by small actuators in the character's slim neck, creates a risk of overheating, amplified by the costume. To keep actuators from overheating, we feed temperature values as additional inputs to policies, introducing new rewards to keep them within bounds. We validate the efficacy of our modeling in simulation and on hardware, demonstrating an unmatched level of believability for a costumed robotic character.

Figures (7)

• Figure 1: Olaf Robot.
• Figure 2: Mechatronic Design and RL-based Control. We separate the articulated backbone from the show functions. The backbone is controlled via policies conditioned on the high-level control input $\bm{g}_t$ and trained using a combination of imitation, overheating, and impact rewards. During training, the control inputs are randomized, whereas at runtime, the Animation Engine generates control inputs from puppeteering commands.
• Figure 3: Mechatronic Design. Shells and skirt have been cut away to show the interior. Note that the costume is not shown.
• Figure 4: Path Frame. Visualization of the path-frame concept from grandia2024bdx and the robot's center of mass.
• Figure 5: Thermal Model. Validation of the thermal model in \ref{['eq:temp_model']}, comparing predicted actuator temperatures $T$ with measured values over a 10min rollout. Temperatures are quantized as reported by the actuator.