MotionDisco: Motion Discovery for Extreme Humanoid Loco-Manipulation

2026-06-04

AI Summary

This work proposes a fully automated, human-demonstration-free approach to discover long-horizon locomotion-manipulation skills for humanoid robots in complex environments from scratch. By integrating large language model–guided evolutionary search with sequential dynamic trajectory optimization and motion pruning strategies, the method efficiently explores feasible whole-body motion sequences. High-precision tracking control is achieved through reinforcement learning. For the first time, this framework autonomously generates and deploys highly challenging, long-horizon task skills without relying on teleoperation or motion retargeting. The approach successfully discovers diverse and effective trajectories in simulation and seamlessly transfers them to real-world humanoid robots, demonstrating robust zero-shot sim-to-real transfer and eliminating dependence on human-provided demonstrations.

Abstract

We present MotionDisco, a framework that discovers contact-rich, long-horizon humanoid loco-manipulation motions from scratch, without relying on teleoperation or motion retargeting from human demonstrations. This is challenging because the space of possible contact interactions grows combinatorially with the task horizon and the number of objects in the scene. MotionDisco enables rapid discovery of novel motions by coupling a large language model (LLM) guided evolutionary search over sequences of interactions with an efficient sequential kinodynamic trajectory optimizer and pruning strategy, enabling the rapid discovery of novel skills. Through extensive ablation studies, we show that our LLM-guided search discovers successful whole-body trajectories across several challenging long-horizon tasks. Finally, by training reinforcement learning tracking policies on the discovered trajectories, we transfer the motions to a real humanoid robot. This is the first work to discover and deploy long-horizon humanoid loco-manipulation skills entirely through automated evolutionary search. Supplementary videos of the experiments are available at: https://youtu.be/DHiVz34QYlw.

Problem

• humanoid locomotion
• manipulation
• contact-rich motion
• long-horizon tasks
• motion discovery

Innovation

• Motion Discovery
• Humanoid Loco-Manipulation
• LLM-Guided Search
• Evolutionary Optimization
• Kinodynamic Trajectory Planning