学术与技术
In this work, we revisit the scaling recipe for BFMs and demonstrate that substantial performance gains can be achieved through the coordination of three core components: 1) the learning paradigm of motion tracking that reformulates diverse humanoid control problems as the reproduction of integrated whole-body behaviors in the global frame; 2) the strategic synergy between on-policy rollout quantity and reference motion diversity; and 3) the expressive and scalable model architecture termed Humanoid Transformer that facilitates the natural emergence of structured behavioral representations.
medium · Score 91.5 · 机器人基础模型
Relative to OpenVLA, CosFly-VLA-0.8B reduces open-loop Average Displacement Error (ADE) by 34.1% on seen-test and 35.3% on unseen-test.
medium · Score 89.2 · VLA
Rather than denoising iteratively at inference, DriftWorld learns an action-conditioned drift during training, allowing it to generate future frames from the current observation and a candidate action sequence in a single forward pass at 30+ fps, which is 17x faster on average than diffusion based baselines.
medium · Score 87.6 · 行业动态
To obtain aligned visual, state, language, and dynamically feasible action data, we build a DiffAero-based pipeline with complementary Isaac Lab and 3D Gaussian splatting renderers.
medium · Score 87.1 · 行业动态
On challenging real-robot manipulation tasks, RoboTTT improves overall performance by 87% over the single-step context baseline and fully completes a five-minute, ten-stage assembly task, which no baseline ever does.
medium · Score 86.9 · 灵巧操作
We are pleased to announce the 22nd edition of the “Robotics: Science and Systems” (RSS) conference to be held at the University of Technology Sydney and the International Convention Centre, Sydney, Australia from July 13-17, 2026.
medium · Score 83.3 · 学术会议与论文
千寻智能（Spirit AI）成立于2024年1月，是一家全球领先的专注于构建机器人“通用大脑”的具身智能公司。公司致力于研发通用具身大模型，赋予机器人跨场景的卓越泛化性与精细的物理交互能力，解决具身智能从虚拟到现实落地的核心难题。作为行业领军者，千寻智能自研的 Spirit 系列模型（如 Spirit v1.5）在 RoboChallenge 全球权威基准测试中持续领跑，代表了当前具身智能模型能力最领先的技术水平。千寻智能凭借卓越的技术实力和创新理念，致力于让通用的机器人伙伴走进千家万户，驱动世界迈向智能机器人时代。
medium · Score 80.1 · 具身大脑生态、机器人基础模型
NVIDIA today announced Project GR00T, a general-purpose foundation model for humanoid robots, designed to further its work driving breakthroughs in robotics and embodied AI.
high · Score 74.9 · 机器人基础模型
开源与工具
Tags: lerobot, safetensors, robotics, pi052, dataset:pepijn223/robocasa_pretrain_human300_v4_annotated5, license:apache-2.0, region:us
high · Score 90.1 · 开源模型与数据
## Release Blog https://huggingface.co/blog/lerobot-release-v050 ## What's Changed * chore(dependencies): Bump lerobot to 0.4.5 by @imstevenpmwork in https://github.com/huggingface/lerobot/pull/3051 * chore: add AI policy by @imstevenpmwork in https://github.com/huggingface/lerobot/pull/3055 * Improve policy_device documentation for async.mdx by @btelles in https://github.com/huggingface/lerobot/pull/3060 * fix(frame_index): making rerun's "frame_index" timeline compatible with behaviour1k datasets by @CarolinePascal in https://github.com/huggingface/lerobot/pull/3068 * fix(dataset edit tools): clarifying `root` argument usage + adding related features by @CarolinePascal in https://github.co
high · Score 78.8 · 行业动态
https://github.com/user-attachments/assets/0d2938ff-ee35-4195-ae80-d7dacb50dc2f ## What's Changed * chore(dependencies): Bump lerobot to 0.5.2 by @imstevenpmwork in https://github.com/huggingface/lerobot/pull/3307 * refactor(envs): move benchmark dispatch into EnvConfig subclasses by @pkooij in https://github.com/huggingface/lerobot/pull/3272 * feat(envs): lazy env init + AsyncVectorEnv as default for n_envs > 1 by @pkooij in https://github.com/huggingface/lerobot/pull/3274 * feat(ci): add agent assitance workflow by @imstevenpmwork in https://github.com/huggingface/lerobot/pull/3332 * chore(security): update claude.yml by @hf-security-analysis[bot] in https://github.com/huggingface/lerobot/
high · Score 78.0 · 行业动态
Tags: lerobot, safetensors, lingbot_va, robotics, license:apache-2.0, region:us
high · Score 78.0 · 开源模型与数据
Tags: lerobot, safetensors, lingbot_va, robotics, license:apache-2.0, region:us
high · Score 78.0 · 开源模型与数据
Tags: lerobot, safetensors, lingbot_va, robotics, license:apache-2.0, region:us
high · Score 78.0 · 开源模型与数据
It builds on [v3.0.0-beta](https://github.com/isaac-sim/IsaacLab/releases/tag/v3.0.0-beta) with additional features and improvements on Newton support (VBD, solver coupling, Kamino, rough terrain, sensors), multi-backend physics, simplified training and installation commands, kit-less workflows, visualizers, rendering, teleoperation, learning exports, installation, CI, and documentation.
high · Score 74.8 · 行业动态
Tags: task_categories:other, language:en, arxiv:2601.02078, region:us, real-world, dual-arm, robotics manipulation, simulation
high · Score 74.6 · 灵巧操作、仿真到现实、开源模型与数据
公司与产品
# Isaac Lab 3.0 Beta 2 - Patch 1 This is a small patch release on top of the previous Isaac Lab 3.0.0 Beta 2 release, including an update to support Isaac Sim 6.0.1, which includes fixes and improvements for NuRec workflows (https://docs.isaacsim.omniverse.nvidia.com/6.0.1/overview/release_notes.html).
high · Score 77.5 · 行业动态
Research Blog https://research.nvidia.com/labs/gear/gr00t-n1_6/ Minor Fixes for GR00T 1.6 Release - Removal of dangling references in Dockerfile and pyproject.toml's.
high · Score 77.1 · 机器人基础模型
Research Blog https://research.nvidia.com/labs/gear/gr00t-n1_5/
high · Score 77.1 · 机器人基础模型
Our foundation model, the Skild Brain, which works on various quadrupeds, humanoids, table-top arms, mobile manipulators and more, follows a hierarchical architecture: (1) a low-frequency high-level manipulation and navigation action policy which provides inputs to a (2) high-frequency low-level action policy.
high · Score 75.3 · 机器人基础模型、灵巧操作
In our experiments, we see π0.7 exhibiting the first signs of compositional generalization, recombining skills from various tasks to solve new problems, like using new kitchen appliances and even enabling a new robot to fold laundry for which there is no laundry folding data.
high · Score 73.5 · VLA、行业动态
This is, however, a catch-22 : robots need data to improve, but only capable robots can be deployed to gather it.
high · Score 69.3 · 行业动态
We did not do everything possible for the highest success rate (as discussed, e.g., in our recent work on using RL for optimizing reliability and speed), and the policies for these tasks are often not consistent, though on average they have a success rate of 52% and a task progress of 72%.
high · Score 64.4 · 行业动态