Search papers, labs, and topics across Lattice.
MotionDisco introduces a novel framework for discovering long-horizon humanoid loco-manipulation motions autonomously, without the need for human demonstrations or teleoperation. By integrating a large language model with an evolutionary search and a kinodynamic trajectory optimizer, the system effectively navigates the complex space of contact interactions to generate successful whole-body trajectories. The approach not only demonstrates the feasibility of automated skill discovery but also successfully transfers these motions to a real humanoid robot, marking a significant advancement in robotic manipulation capabilities.
Automated discovery of complex humanoid manipulation skills could revolutionize how robots learn and adapt to new tasks without human input.
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.
CMU ML