Tag
Google DeepMind's Gemini 4 Argon learned to play cornhole by watching a single video, inferred the physics, generated its own simulation, and trained Apptronik's Apollo humanoid to land seven consecutive throws.
WUJI showcases its Hand 2 humanoid dexterous hand, which mirrors human hand movements via camera tracking or a glove and learns skills like pen spinning through reinforcement learning in mjlab simulation before transferring them to the real world. The wuji-mjlab code is released as open source for developers and researchers.
GreenWAM 提出一个机器人世界基础模型，通过共享的解剖学动作空间联合训练 20 个机器人家族，使其既能作为世界模型预测动作后果，又能作为策略控制不同形态的机器人，并在多个基准上取得领先表现。
OTRetarget 提出一种统一方法，利用熵最优传输联合重定向机器人与多物体的运动，通过保持接触一致性显著优于 OmniRetarget，并在物理人形机器人 G1 上验证了迁移到强化学习训练的全身策略的可行性。
Figure released Helix 2.5, a humanoid control model that performed household tasks zero-shot across 30 homes using pretrained data.
Ati Robotics, an India-based startup, built robots with minimal Chinese parts in response to US regulatory bans on Chinese-made robots. Founder Saurabh Chandra discusses their supply chain strategy, cost advantages, and upcoming humanoid robot.
DeepMind introduces Gemini Robotics 2, a set of AI models (VLA, ER, On-Device) that enable robots to perform whole-body control, fine dexterity, and multi-robot collaboration, with local on-device execution and fast adaptation to new robot bodies.
Alexander Koch released the first open-source robot arm integrated with Hugging Face's LeRobot platform, and announced Tau's humanoid cleaning service in San Francisco at $30/hour.
Moby, a humanoid robot, demonstrates powerful whole-body control for handling dynamic payloads, marking significant progress over current humanoids.
A new personal robot called the T1, showcased at the AI for Good Summit in Geneva, can transform between a four-legged dog-like shape and a humanoid form, allowing it to navigate both smooth and rough terrain, climb steps, and follow users.
UMA_Robots unveils Northstar, a robot with a human-like body and hands, designed to be approachable and competent for workplaces and homes.
A tweet promotes an interactive website that breaks down humanoid robot hardware, costs, and supply chains, highlighting materials as the bottleneck.
SceneBot is a unified motion tracking framework that conditions a single humanoid policy on both reference motions and per-link contact labels, enabling free-space locomotion, terrain traversal, and whole-body manipulation in contact-rich environments.
LeRobot released the largest open-source humanoid teleoperation dataset, re-encoded into the LeRobot format for efficient streaming and reduced storage footprint.
This paper introduces DeepInsight, a unified evaluation infrastructure for Physical AI stacks that spans from foundation model decoding to whole-body control, preserving heterogeneity through three narrow abstractions to enable cross-layer diagnostics.
OASIS is a simulation-data-driven framework for humanoid loco-manipulation that uses 3D generative models and hierarchical visuomotor policies. It achieves better zero-shot performance than real-robot training by leveraging domain randomization in simulation.
This paper introduces LIMMT, a data-centric study showing that training with high-quality, minimal subsets of motion data (under 3% of AMASS) outperforms using the full dataset for physics-based humanoid motion tracking, defining motion data quality through physics feasibility, diversity, and complexity.
Humanoid-GPT is a GPT-style Transformer pre-trained on a billion-scale motion corpus, achieving zero-shot generalization for whole-body motion tracking across unseen motions and tasks.
Hugging Face has released LeRobot Humanoid, an open-source, low-cost ($2.5k) 3D-printed humanoid robot designed for robot learning, including hardware, software, simulation, and training tools.
LeRobot Humanoid is an open, low-cost bipedal robot platform for robot learning, including hardware, runtime, identification tools, and training environments, built from 3D-printed and off-the-shelf parts for about $2,500.