Researchers are exploring new methods for controlling humanoid robots, focusing on advanced loco-manipulation and reactive behaviors. One approach uses diffusion models to generate functional neural networks for robots like the Unitree G1, analyzing the learned weight space for insights into embodiment. Another framework, Reflex, enables humanoids to visually react to dynamic objects by inferring object motion from egocentric vision and coordinating whole-body control. Additionally, HULK is a whole-body control framework that uses model predictive control to guide reinforcement learning for forceful loco-manipulation, improving balance and command tracking. Finally, a new dataset and pipeline called HumanVerse are introduced to scale human demonstrations for training humanoid foundation models, enabling more effective whole-body interaction. AI
IMPACT Advances in humanoid robot control could accelerate their deployment in complex physical tasks and human environments.
RANK_REASON Multiple research papers detailing new methods and datasets for humanoid robot control.
- Reflex
- Unitree G1
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Diffusion Transformer
- Gotit.pub
- Hugging Face
- HumanVerse
- imitation learning
- reinforcement learning
- ScienceCast
AI-generated summary · Google Gemini · from 4 sources. How we write summaries →
