MotionDiscoMotion Discovery for Extreme Humanoid Loco-Manipulation

Ilyass Taouil1,*,Michal Ciebielski1,*,Shafeef Omar1,*,
Haizhou Zhao1,2,Angela Dai1,Aaron M. Johnson3,Majid Khadiv1
1Technical University of Munich, Germany, 2New York University, USA, 3Carnegie Mellon University, USA
* Equal contribution
arXivVideo

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.

Overview


Overview of the MotionDisco framework: an LLM-guided evolutionary tree search over interaction programs coupled with a contact-explicit motion planner.

MotionDisco couples LLM-guided evolutionary discovery of contact plans (left) with contact-explicit trajectory optimization (right). Each search node proposes a mutation of its parent program, conditioned on the goal prompt and the parent’s feasibility feedback; executing the mutated program yields a discrete contact plan (denoted by the blue dots). Plans that pass a kinematic feasibility check are sent to the trajectory optimizer, which assesses dynamic feasibility and returns the feedback that guides subsequent mutations.

Discovered Skills for
Diverse Long-Horizon Tasks

01 / 06

Reach the banana hanging from the ceiling

02 / 06

Climb on the table with the box

03 / 06

Climb on the table and place the object on the higher table

04 / 06

Pick and place over long distance

05 / 06

Move through clutter to the other side

06 / 06

Pick the box from under the table and place it on top

More Simulation Results

Long-horizon reasoning and motion generation for more tasks.

Citation

BibTeX
@misc{taouil2026motiondiscomotiondiscoveryextreme,
      title={MotionDisco: Motion Discovery for Extreme Humanoid Loco-Manipulation}, 
      author={Ilyass Taouil and Michal Ciebelski and Shafeef Omar and Haizhou Zhao and Angela Dai and Aaron M. Johnson and Majid Khadiv},
      year={2026},
      eprint={2606.06139},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2606.06139}, 
}