Sumo: Dynamic and Generalizable Whole-Body Loco-Manipulation

Sumo combines a pre-trained whole-body control policy with sample-based planning at test time, enabling legged robots to perform dynamic loco-manipulation without retraining for each new task. The approach generalizes across objects and objectives, including real-world demonstrations on a Spot quadruped manipulating heavy and oversized objects, and simulated humanoid tasks such as opening doors and pushing tables.

• Website
• Paper
• Code
• Video

Related Papers

2026 April

Sumo: Dynamic and Generalizable Whole-Body Loco-Manipulation John Zhang, Maks Sorokin, Jan Bruedigam, Brandon Hung, Stephen Phillips, Dmitry Yershov, Farzad Niroui, Tong Zhao, Leonor Fermoselle, Xinghao Zhu, Chao Cao, Duy Ta, Tao Pang, Jiuguang Wang, Preston Culbertson, Zac Manchester, and Simon Le Cleac'h arXiv (In Review)

People

John Zhang

Optimization, Control, and Robotics

Jan Bruedigam

M.S. Visitor, Stanford. Now at RAI Institute.

Brandon Hung

M.S. in ECE, CMU. Now at RAI Institute.

Zac Manchester

Associate Professor

Simon Le Cleac'h

Ph.D. in MechE, Stanford. Now at RAI Institute.

Last updated: 2026-04-09