Engineering & computing · OECD FoS 2.2 · China MoE 0811 控制科学与工程
Robotics & automation
机器人与自动化
L2Local breakthroughs ↑ Rising Score 2.20 · Last reviewed 2026-10-22 · Next review 2027-01-19
BIGGEST BREAKTHROUGH 2026-07-30 AI · majorField · major
Gemini Robotics 2: whole-body humanoid control, dexterity and multi-robot teamwork ↗
Extends VLA to whole-body humanoids
Why this level
Vision-language-action models give robots some generalisation (RT-2 2023, π0 2024, Gemini Robotics 2025), but real-world reliability and deployment at scale remain unsolved.
S = 0.30·method + 0.35·discovery + 0.25·adoption + 0.10·education (education scores 0 when there is no evidence; if empty, the other weights are renormalised)
Sources
- deepmind.google/discover/blog/rt-2-new-model-translates-vision-and-language-into-action
- deepmind.google/discover/blog/gemini-robotics-brings-ai-into-the-physical-world
Composite 1.5–2.49: at least one peer-reviewed/officially verified AI-key result, or one dimension at 3; routine workflow largely unchanged. All levels →
Research directions
VLA foundation modelsHumanoidsSim-to-realEmbodied datasets
Papers
- $π_{0.7}$: a Steerable Generalist Robotic Foundation Model with Emergent Capabilities
Out-of-the-box performance near specialised RL-finetuned models
2026-04-16 · arXiv
- Open X-Embodiment: Robotic Learning Datasets and RT-X Models
Foundation dataset for cross-embodiment policies
2023-10-13 · arXiv
- Agentic World Action Models
AW-0 couples code-based planning with joint video and action prediction for compositional robot manipulation.
2026-10-21 · arXiv · Zhuoyang Liu, Dantong Niu, Botong Zhang, Boning Shao
- AtomicRL: Atomic Reinforcement Learning for Long-Horizon Robotic Manipulation
Atomic subtask-level RL for VLA policies tackles sparse, biased credit and uneven subtask mastery in long-horizon manipulation.
2026-10-21 · arXiv · Likui Zhang, Zhe Zhu, Zhihao Zhan, Yibo Bai
- UniWALM: Unified World Action Language Model for Embodied Robot Control
A single backbone co-generating video, action and language, using generated language as memory for robot control.
2026-10-20 · arXiv · Rahul Sajnani, Diego Romeres, Mingyuan Zhang, Kai Gao
- AgenticRoboRL: Bringing Coding-Agent Post-Training to Agentic Robotics
Post-trains robot planners with SFT and multi-turn RL across harnesses, improving a 9B planner on LIBERO-PRO.
2026-10-20 · arXiv · Xinxin Liu, Bingxin Xu, Wenrui Bao, Ming Li
- CoGaussian: Open-Vocabulary Gaussian-based Collaborative 3D Scene Understanding
Foundation-model-aligned sparse 3D Gaussians for open-vocabulary collaborative occupancy across agents.
2026-10-20 · arXiv · Om Shivam Verma, Shaya Farahmand, Lingjun Zhao, Lu Gan
- Ropedia W0: Scaling Loco-Manipulation World Models across Embodiments with Human Experience
First egocentric loco-manipulation world model across humans and robots, with 10k h human and 3k h robot data.
2026-10-19 · arXiv · Tianqi Liu, Hanfei Zhao, Qifeng Liang, Hui En Pang
- Physical Words: A Vocabulary for Structuring Latent Reasoning in Robot Manipulation
PhysWord structures latent physical reasoning with atomic action verbs before acting.
2026-10-19 · arXiv · Jiaming Liu, Xinghao Zhu, Yiming Zhang, Shanghang Zhang
- GPT-6 Astra Is a Native Visuo-Tactile Policy for Contact-Rich Manipulation
Claims GPT-6 Astra directly interprets visuo-tactile images for few-shot contact-rich manipulation on 19 simulated tasks.
2026-10-19 · arXiv · Chen Yang, Linzhe Shi, Hang Zhang, Mu Xu
- From arXiv, bioRxiv, medRxiv and OpenAlex (last ~30 days); summaries written by AIMap from the abstract.
- Ropedia W0: Scaling Loco-Manipulation World Models across Embodiments with Human Experience 2026-10-19
- Learning from Fleet Experience: Collective Experience Learning Across 96 Robots 2026-10-19
- ScaleNav: Scalable Internet Video Pretraining for Language-Conditioned Local Navigation 2026-10-19
- VLA-Rolling: Fewer Denoising Steps, Better Control via Adaptive Action Reuse 2026-10-18
- HexaShell: A Thin Harness for Exploring the Limits of VLM Agents as Robot Coding Agents 2026-10-16
- Has GPT-6-Astra Achieved Embodied Intelligence? 2026-10-15
- All related papers in the archive →
Milestones 8
Big glowing nodes are leaps: AI leap cool · Field leap warm · double ring = both
-
Industry
Physical Intelligence opens Pi API beta for hosted fine-tuning and inferenceRobot foundation models offered as a service
Physical IntelligenceVendor claim -
Regulatory
Korea’s regulatory sandbox approves 20 Robros AI humanoids to work alongside people in semiconductor, auto, shipbuilding and logistics sitesAI: robots must perceive and manipulate in human-shared spaces. Field: first government safety exemption for humanoids working beside people in real plants
Seoul Economic Daily -
System
Gemini Robotics 2: whole-body humanoid control, dexterity and multi-robot teamworkExtends VLA to whole-body humanoids
Google DeepMindVendor claim -
System
π0.7: steerable generalist robot foundation model with zero-shot cross-embodimentOut-of-the-box performance near specialised RL-finetuned models
arXiv / Physical IntelligenceVendor claimPreprint -
System
π0 generalist robot policyGeneralist policy for long-horizon dexterous tasks
Physical Intelligence -
Dataset
Open X-Embodiment: 22 robot types, 1M+ trajectories, RT-X modelsFoundation dataset for cross-embodiment policies
arXivPreprint -
System
RT-2 vision-language-action modelFirst VLA transferring web knowledge to actions
Google DeepMind
No milestones match this filter.
More papers & repos 3
| Type | Title | Date |
|---|---|---|
| Paper | $π_{0.7}$: a Steerable Generalist Robotic Foundation Model with Emergent Capabilities Preprint | 2026-04-16 |
| Paper | Open X-Embodiment (arXiv 2310.08864) PreprintTo be verified | 2023-10 |
| Repo | Physical-Intelligence/openpi To be verified | — |
Latest news 12
- Gecko Robotics secures $300M contract with the U.S. NavyThe Robot Reporten
- Tennant Company Introduces Its First Autonomous Robotic Vacuum, the P1 VACRobotics Tomorrowen
- 2027 TRR50 Robotics Innovation Awards open nominationsThe Robot Reporten
- 优必选成立具身智能公司，注册资本2000万元36氪zh
- Pi API betaPhysical Intelligenceen
- Gemini Robotics 2 brings whole body intelligence to robotsGoogle DeepMind Blogen