Unleashing Infinite Motion: Scaling Expressive Quadrupedal Motion via Generative Video Priors

2026-06-26

AI Summary

This work addresses the limitations of current quadrupedal robots, which exhibit limited behavioral diversity and rely heavily on real animal data, suffering from fragile cross-species motion reconstruction and morphological adaptation challenges. To overcome these issues, the authors propose Uni-Mo, a fully automated framework that reframes the scarcity of motion data as a generative task. Leveraging large language models to produce semantic behavior prompts, the framework synthesizes identity-consistent quadruped motion videos via a video diffusion model and extracts 3D trajectories to train reinforcement learning-based tracking policies. The study introduces Quad-Imaginarium, an open-source dataset comprising 7,488 annotated sequences (18.5 hours total). Evaluated on the Unitree Go2 platform, the method successfully deployed 392 randomly selected motions with a 96.7% real-world success rate and 97.6% in simulation, substantially enhancing the expressiveness and generalization capabilities of quadrupedal robots.

Abstract

Quadruped robots have achieved remarkable locomotion, yet their behavioral repertoire remains confined to a few gaits--far from the expressive, companion-like presence long envisioned for them. Attempts to import the humanoid recipe of large-scale motion data have inherited one tacit assumption: that robot motion must first pass through an animal body, making data collection dependent on cooperative animals, reconstruction fragile across species, and retargeting ill-posed across incompatible morphologies. We propose Uni-Mo, a fully automated pipeline that removes the animal from the loop by reframing data scarcity as a generation problem: an LLM proposes motion prompts, a video diffusion model synthesizes the corresponding robot behaviors, and the generated videos are lifted into 3D reference trajectories used to train tracking policies deployed on a real Unitree Go2. To make naively-drifting generations reliably extractable, we introduce an Identity Consistency Loss that enforces appearance coherence across frames. We release Quad-Imaginarium at https://github.com/GaoLii/Quad-Imaginarium.git, the resulting open-source dataset of 7,488 language-annotated quadruped motions (18.5 hours) spanning acrobatic and performative behaviors. We validate 392 randomly sampled motions on a real Unitree Go2 with a 96.7% deployment success rate, complemented by a 97.6% success rate across the full dataset in simulation.

Problem

• quadrupedal motion
• motion generation
• data scarcity
• behavioral repertoire
• motion retargeting

Innovation

• generative video priors
• quadrupedal motion generation
• identity consistency loss
• motion retargeting
• diffusion models