HEFT says a full-size humanoid can teleoperate with payloads up to 24 kg

A new robotics preprint says a full-size humanoid can be trained to teleoperate while carrying real payloads, a capability that has been difficult on larger machines because of inertia and tighter balance margins.

The paper, titled "HEFT: Heavy-Payload Full-size Humanoid Teleoperation with Privileged Motion Guidance and Windowed Payload Curriculum," was posted on arXiv on July 2, 2026 at 15:37:37 UTC. The authors are Chenxin Liu, Qingzhou Lu, Guangxiao Yang, Xuanyang Shi, Chenghan Yang, Yanjiang Guo, and Jianyu Chen.

What HEFT Claims

HEFT is presented as a teleoperation framework for full-size humanoids that combines two training ideas. The first, Privileged Motion Guidance (PMG), uses reconstructed motion during training while the system is driven by raw VR references. The second, Windowed Payload Curriculum (WPC), gradually expands the payload range with expert-guided caps.

In plain terms, the paper is trying to bridge the gap between noisy human teleoperation inputs and the more precise motion targets needed to train a large humanoid robot under load.

Robot And Demonstrations

The authors say they deployed the system on L7, a humanoid robot that is 175 cm tall, weighs 65 kg, and has 29 actuated joints.

According to the preprint, the robot successfully tracked turns, forward and backward locomotion, and squats while carrying real two-hand loads. The paper says the payload ceiling reached 24 kg.

That matters because the challenge is not just moving the robot through a demo motion. On a full-size platform, adding weight changes the balance problem and makes tracking more sensitive to tracker noise and other teleoperation errors.

Why It Matters

The authors position heavy-payload teleoperation as an underexplored path for humanoid skill acquisition, especially for robots meant to operate at human scale in physical environments.

If the results hold up beyond the preprint, the approach could broaden the range of tasks humanoids can learn from teleoperation, moving beyond unloaded imitation toward heavier manipulation and locomotion work.

What To Watch Next

The current evidence is from a preprint, so it has not yet been independently validated through peer review.

The next things to watch are whether the authors release a project page, videos, or code, and whether the work appears in a conference or journal submission timeline. It will also be useful to see how the 24 kg result compares with prior full-size humanoid teleoperation systems on similar motions.