# HULK framework lets humanoid robots lift heavy loads while staying balanced

> Researchers released a whole‑body control system that reduces balance errors by 35.7% and tolerates torso pushes of up to 130 N.

Oossa · 2026-10-08 · https://oossa.com/en/hulk-framework-lets-humanoid-robots-lift-heavy-loads-while-staying-balanced

A team led by An Dang introduced HULK, a control framework for humanoid robots that need to move while holding large objects. The system combines model predictive control with reinforcement learning and a balance‑keeping barrier function. Tests in simulation and on a Unitree G1 robot showed lower tracking errors and better stability when each arm carried 10 kg. The wrist‑force teacher could resist pushes on the torso up to 130 N.

## The facts

- 10 kg per arm payload tested
- 35.7% reduction in DCM excursion compared with MPC‑only learning

## Why it matters

It lets robots safely transport heavy items in settings like warehouses or disaster sites.

## Sources & references

1. [HULK: Learning Whole-Body Forceful Loco-Manipulation for Humanoids](https://arxiv.org/abs/2610.08970) – arXiv cs.RO, 2026-10-08

Last updated: 2026-10-08