A multi-source 16,074-clip quadruped motion library plus a flow-matching generalist tracker shows empirical data scaling and zero-shot unseen tracking, integrated with all-terrain locomotion and real-robot deployment.
From experts to a generalist: Toward general whole-body control for humanoid robots
8 Pith papers cite this work. Polarity classification is still indexing.
8
Pith papers citing it
abstract
Achieving general agile whole-body control on humanoid robots remains a major challenge due to diverse motion demands and data conflicts. While existing frameworks excel in training single motion-specific policies, they struggle to generalize across highly varied behaviors due to conflicting control requirements and mismatched data distributions. In this work, we propose BumbleBee (BB), an expert-generalist learning framework that combines motion clustering and sim-to-real adaptation to overcome these challenges. BB first leverages an autoencoder-based clustering method to group behaviorally similar motions using motion features and motion descriptions. Expert policies are then trained within each cluster and refined with real-world data through iterative delta action modeling to bridge the sim-to-real gap. Finally, these experts are distilled into a unified generalist controller that preserves agility and robustness across all motion types. Experiments on two simulations and a real humanoid robot demonstrate that BB achieves state-of-the-art general whole-body control, setting a new benchmark for agile, robust, and generalizable humanoid performance in the real world. The project webpage is available at https://beingbeyond.github.io/BumbleBee/.
years
2026 8representative citing papers
ExoActor uses exocentric video generation to implicitly model robot-environment-object interactions and converts the resulting videos into task-conditioned humanoid control sequences.
TeleGate achieves high-precision real-time whole-body teleoperation of humanoid robots by dynamically gating between expert policies and using a VAE motion prior to infer future intent from history, outperforming distillation baselines on dynamic motions with only 2.5 hours of mocap data.
Dual-expert RL plus difficulty-aware multi-teacher distillation improves physics-based human–scene interaction imitation under complex 3D geometry versus prior single-policy baselines.
Unsupervised skill mining with self-supervised compactness, alignment, and disentanglement losses yields a fixed skill library that improves multi-task and few-shot robotic manipulation when plugged into ACT and OpenVLA-OFT.
Humanoid-GPT is a causal Transformer pre-trained on a unified billion-scale motion dataset that tracks dynamic behaviors with zero-shot generalization to unseen motions and tasks.
HoloMotion-1 trains a MoE Transformer policy on hybrid video and MoCap motion data to achieve robust zero-shot tracking that transfers directly to real humanoid robots.
Switch enables humanoid robots to perform agile, seamless transitions between locomotion skills via a kinematic skill graph, DRL tracking policy, and real-time graph-search scheduler.
citing papers explorer
-
Behavior Foundations for Quadruped Robots: ABot-C0 Technical Report
A multi-source 16,074-clip quadruped motion library plus a flow-matching generalist tracker shows empirical data scaling and zero-shot unseen tracking, integrated with all-terrain locomotion and real-robot deployment.
-
ExoActor: Exocentric Video Generation as Generalizable Interactive Humanoid Control
ExoActor uses exocentric video generation to implicitly model robot-environment-object interactions and converts the resulting videos into task-conditioned humanoid control sequences.
-
TeleGate: Whole-Body Humanoid Teleoperation via Gated Expert Selection with Motion Prior
TeleGate achieves high-precision real-time whole-body teleoperation of humanoid robots by dynamically gating between expert policies and using a VAE motion prior to infer future intent from history, outperforming distillation baselines on dynamic motions with only 2.5 hours of mocap data.
-
ComplexMimic: Human-Scene Interaction Imitation in Complex 3D Environments
Dual-expert RL plus difficulty-aware multi-teacher distillation improves physics-based human–scene interaction imitation under complex 3D geometry versus prior single-policy baselines.
-
SkillPlug: Unsupervised Skill Mining for Few-Shot Adaptation in Robotic Manipulation
Unsupervised skill mining with self-supervised compactness, alignment, and disentanglement losses yields a fixed skill library that improves multi-task and few-shot robotic manipulation when plugged into ACT and OpenVLA-OFT.
-
Humanoid-GPT: Scaling Data and Structure for Zero-Shot Motion Tracking
Humanoid-GPT is a causal Transformer pre-trained on a unified billion-scale motion dataset that tracks dynamic behaviors with zero-shot generalization to unseen motions and tasks.
-
HoloMotion-1 Technical Report
HoloMotion-1 trains a MoE Transformer policy on hybrid video and MoCap motion data to achieve robust zero-shot tracking that transfers directly to real humanoid robots.
-
Switch: Learning Agile Skills Switching for Humanoid Robots
Switch enables humanoid robots to perform agile, seamless transitions between locomotion skills via a kinematic skill graph, DRL tracking policy, and real-time graph-search scheduler.