← Paper Digest

Awaken-HI: Scaling Humanoid Intelligence with Whole-Body Human Experience

EngineAI

November 2026

Bimanual Object Storage: the T800 reaches for a cup and a toy car to put them in bins.

TL;DR EngineAI records 50,000 hours of egocentric stereo video with synchronized whole-body human motion and pretrains a humanoid VLA to predict both the original human motion and the retargeted robot actions. After post-training on 150 hours of teleoperation it averages 93.75% success on eight real T800 loco-manipulation tasks, against 57.5% for π0.5 and 58.75% for GR00T N1.7.

What they do

Whole-body teleoperation is too costly to cover many humanoid tasks, and egocentric human video gives no whole-body motion targets. The authors record Awaken-Human: about 50,000 hours of egocentric stereo video with synchronized whole-body motion, covering more than 100 skills in homes, workplaces, shops and storage areas. They pretrain a VLA, Awaken-HI, on it, then post-train it on the EngineAI T800 humanoid.

How it works

  • Capture and annotation. A PICO headset and five body trackers record stereo video and skeletons. An in-house 3D hand model replaces the PICO hand tracking, stereo matching supplies depth, and a VLM proposes subtask labels that people correct.
  • Retargeting. A T800-specific pipeline built on GMR cuts mean end-effector position error from 16.4 cm to 2.4 cm.
  • Shared action expert. A Qwen3-VL backbone feeds one flow-matching diffusion transformer with three branches: retargeted 40-D robot actions, the original 161-D human motion and camera-frame wrist poses. Extra heads predict current and future depth and the current subtask.
  • Post-training on about 150 hours of T800 teleoperation, using only the robot-action loss.

Results

  • On the real T800 (10 trials per task), Awaken-HI averages 93.75% over eight tasks, against 57.5% for π0.5 and 58.75% for GR00T N1.7. On the long-horizon industrial and household tasks it reaches 90% and 80%, while neither baseline exceeds 30%.
  • Growing pretraining data from 5,000 to 50,000 hours lowers held-out action MSE by about 11% (power-law exponent 0.051). Mean success on three household tasks rises from 56.7% to 90.0%.
  • In the 5,000-hour ablation, adding human-motion supervision lifts household success from 13.3% to 43.3%, and all auxiliary targets together reach 53.3%.
  • New object-destination pairings succeed 80-100% of the time, against 0-40% for the baselines.

Limitations

The paper has no limitations section. All results come from one humanoid with ten trials per condition, and the scaling and ablation robot tests cover only three household tasks. Transfer to unseen objects is partial: 50% completion for tissues and 30% for a medicine bottle, though both baselines score 0%.

Videos

Garment Hanging: the T800 picks up a garment from a table.
Laundry Basket Relocation: the T800 works at a wardrobe to retrieve a laundry basket.

Figures

From Figure 1: Overview of Awaken-HI and the Awaken-Human dataset. Awaken-Human pairs approximately 50,000 hours of e...
From Figure 1: Overview of Awaken-HI and the Awaken-Human dataset. Awaken-Human pairs approximately 50,000 hours of egocentric stereo video with synchronized whole-body motion (left). We pretrain Awaken-HI on human motion and retargeted robot actions, then fine-tune it on robot demonstrations (center). The right panels summarize scaling and generalization results alongside the full model’s 93.75% mean success rate across eight individual tasks.
From Figure 2: Object and skill coverage in Awaken-Human. The dataset contains 17.02 million episodes spanning more t...
From Figure 2: Object and skill coverage in Awaken-Human. The dataset contains 17.02 million episodes spanning more than 100 skills. The left panels show the distribution of object categories (top) and episode counts for selected skills (bottom). The right panel shows representative egocentric frames.
From Figure 4: Two-stage training of Awaken-HI. We pretrain Awaken-HI on Awaken-Human, using a shared action expert t...
From Figure 4: Two-stage training of Awaken-HI. We pretrain Awaken-HI on Awaken-Human, using a shared action expert to jointly predict human motion, wrist poses, and retargeted robot actions, with auxiliary depth and subtask supervision. We then retain the shared expert and robot action branch and fine-tune the policy on real-robot demonstrations.
From Figure 6: Scaling whole-body egocentric pretraining improves action prediction and real-robot success. From left...
From Figure 6: Scaling whole-body egocentric pretraining improves action prediction and real-robot success. From left to right, the panels show inference MSE, validation loss, empirical power-law fits for both metrics, and real-robot success rates after post-training on three household tasks. Their mean success rises from 56.7% to 90.0% as pretraining data grows from 5,000 to 50,000 hours.
Abstract

A central goal of humanoid robotics is for robots to perform the everyday work people do in homes and workplaces. Existing egocentric video datasets cover many of these activities but provide little synchronized whole-body motion data for learning loco-manipulation at scale. We build Awaken-Human, which pairs approximately 50,000 hours of egocentric stereo video with synchronized whole-body human motion and covers more than 100 skills. Building on this, we also introduce Awaken-HI (Humanoid Intelligence), a vision-language-action model for humanoid loco-manipulation that jointly predicts original human motion and retargeted robot actions through a shared action expert. We show that loco-manipulation skills transfer more effectively from people to humanoids when human trajectories directly supervise the shared action expert. Increasing human pretraining data reduces held-out robot action prediction error according to an empirical scaling law and raises the mean success rate on downstream real-robot tasks. After full-corpus pretraining and T800 adaptation, Awaken-HI achieves a mean success rate of 93.75% across eight individual tasks and outperforms GR00T N1.7 and $π_{0.5}$ in both individual and long-horizon evaluations. The same policy generalizes to objects unseen during robot post-training and to new pairings of familiar objects and destinations.

Summaries are written automatically from each paper's text. Figures and videos come from the paper and its project page; all credit goes to the authors.