Robotics & Embodied AI
Physical embodiment, VLA models, manipulation, navigation, and real-world interaction. AI that can move and interact with the physical world.
Key Benchmarks
Recent Papers
VISTA: A Visual Harness for Reasoning in an Interactive World
Qiushi Han, Keya Hu, Linlu Qiu +2 more
VABench: Measuring Embodied Spatial Intelligence through Visual Demonstrations, Active Perception, and Metric Control
Multiple authors
Embodied-BenchForge: A Closed-Loop Agentic Workflow for Embodied Benchmark Construction
Baoyang Jiang, Fengchun Zhang, Leyuan Wang +9 more
RoboTok: An Internet-Scale Data Engine for Human Demonstration Retrieval and Dexterous Manipulation Learning
Howard Qian, Yiting Chen, Yunfei Xie +6 more
Puffin-World: Scaling a Unified Multimodal Model with Native 3D World States
Kang Liao, Yihang Luo, Xiao-Ming Wu +7 more
PonderPounce: A Pretrained MLLM as an Episode Context Engine for Robot Control
Suhwan Choi, Jaeyoon Jung, Sungkyung Kim +2 more
When Does On-Policy Interaction Help? Representational Tradeoffs in Value-Based Imitation Learning
Luca Viano, Antoine Moulin, Audrey Huang +3 more
ReferTrack: Referring Then Tracking for Embodied Visual Tracking
Hanjing Ye, Tianle Zeng, Jiazhao Zhang +6 more
Self-Supervised Learning of Structured Dynamics from Videos
Lukas Knobel, Andrew Zisserman, Yuki M. Asano
Valdi: Value Diffusion World Models
Christopher Lindenberg, Kashyap Chitta
Recent Milestones
EU EMERGE shows 'awareness' in robot swarms
On October 4, 2026, Scienmag reported on the conclusion of Europe’s EMERGE project, a four‑year EIC‑funded effort combining AI, robotics and cognitive science to study ‘awareness’ in swarms of simple robots. The consortium built theoretical frameworks and robotic prototypes showing how distributed AI systems can exhibit coordinated, context‑sensitive behaviour without any central controller.
Microduck makes RL robots a mass product
On August 31, 2026, AI daily FAV0 reported that Microduck, a $399 open‑source biped robot built by Pollen Robotics in partnership with Hugging Face, generated over $2.5 million in orders within 24 hours of launch. The product pairs a high‑fidelity simulator with a small reinforcement‑learning‑ready robot, prompting enthusiasts to dub it "the iPhone of robotics."
PsiBot gets $100M+ for embodied AI robots
Chinese embodied AI startup PsiBot has completed a new funding round of more than US$100 million, adding a slate of industrial and financial investors including Ningbo Tuopu Group, a Chery Holdings backed fund, Lens Technology, 37 Interactive Entertainment, Wuhu Investment Holding Group and Fosun Fortune Capital, with Zhuhai Technology Industry Group reinvesting. The capital will support PsiBot’s dual system embodied AI architecture, pairing its Psi R2 operation policy model with the Psi W0 world model for logistics and advanced manufacturing deployments. ([laplaceatlas.com](https://www.laplaceatlas.com/stories/embodied-ai-startup-psibot-raises-over-100-million-in-new-funding-round-57a04125))
XPeng Bets Big: $900M for Humanoid Robots
On August 28, 2026, TechCrunch reported that major Chinese automakers including XPeng, Chery’s AiMOGA unit and BYD are investing heavily in humanoid robots as a future profit center, echoing Tesla’s Optimus strategy. The piece follows XPeng’s announcement earlier in the week that its Dogotix robotics unit raised over $900 million at a $6.3 billion valuation to accelerate its IRON humanoid robot program.
Anthropic debuts Model Hardware Standard for robots
On August 28, 2026, UK outlet Resultsense reported on Anthropic’s new Model Hardware Standard (MHS), a research preview specification that lets AI agents discover and operate lab and factory equipment through a common driver. MHS aims to cut integration time from weeks to hours and is being tested with partners like QuEra, Carnegie Mellon, Genentech and robotics vendors. ([resultsense.com](https://www.resultsense.com/news/2026-08-28-anthropic-model-hardware-standard/))
China’s AI models and robots scale up fast
At a 20 July 2026 press briefing summarised by Sina’s AI hourly report, China’s Ministry of Industry and Information Technology said over 30% of large manufacturing firms now use AI technologies and that global downloads of Chinese open‑source large models have surpassed 10 billion. Separately, official data cited in the same digest said China now has more than 400 humanoid robot models—over half of the global total—and that Chinese four‑legged robots account for nearly 70% of worldwide unit sales.
Japan–Nvidia Build ¥1T Vera Rubin AI Factory
Nvidia CEO Jensen Huang used a two‑day visit to Tokyo on July 15–16 to unveil plans for a “Vera Rubin” national AI factory in partnership with Japan’s Noetra sovereign AI consortium. TechCrunch reports Japan will commit up to ¥1 trillion (~$6.2 billion) over five years to physical‑AI infrastructure and a robotics coalition built on Nvidia’s Cosmos models, with the story published July 19, 2026 (US time).([techcrunch.com](https://techcrunch.com/2026/07/19/what-to-watch-for-after-jensen-huangs-japan-visit))
China scales humanoids toward 100k units/year
Reporting from WAIC 2026 in Shanghai, The Eastern Herald writes that China produced about 40,000 humanoid robots in the first half of 2026, matching its total output for 2025. Robotics firm AgiBot and the Shanghai National Humanoid Robotics Innovation Center told attendees China could exceed 100,000 humanoid units by year-end, with multiple product lines now in pilot deployments. ([easternherald.com](https://easternherald.com/2026/07/19/china-humanoid-robots-production-world-ai-conference/))
LimX Raises Pre‑IPO War Chest for Humanoids
Chinese humanoid robotics startup LimX Dynamics announced on July 14, 2026 that it closed a pre‑IPO round of about RMB 13.6 billion (nearly $2 billion), valuing the company at roughly RMB 15 billion. CryptoBriefing and 36Kr report that the round caps roughly $400 million raised in six months to fund large‑scale deployment of its Luna, Oli and TRON humanoid platforms and an agentic OS.
SceneSmith Uses GPT‑5.2 Agents to Train Robots
On July 13, 2026, MIT CSAIL researchers introduced SceneSmith, a system of three vision‑language model agents that automatically generate detailed 3D indoor environments for robot training. The agents use GPT‑5.2 to design, critique and orchestrate virtual scenes that can be loaded directly into physics simulators.
General Intuition Raises $320m for Game‑Trained World Models
On July 13, 2026, Dutch startup General Intuition disclosed a $320 million Series A round at a $2.3 billion valuation, completed in January and announced in June. The company trains AI models on billions of gameplay clips and will use the funds for research, model development, compute and hiring.
Tripo AI Raises $150M for 3D World Models
On July 2, 2026, Tripo AI announced a $150 million Series A3 round backed by automotive, gaming, internet and tech investors. The San Francisco–based AI 3D and world‑model company said it will use the capital to deepen its 3D foundation models, world‑model research and global commercialization.
UBTECH Pushes Humanoids into China’s Homes
Chinese robotics firm UBTECH has unveiled its full-size UWorld U1 humanoid robot series for consumers, reporting over 10,000 pre-orders following a launch event in Shenzhen on June 30, 2026. The three-model lineup, priced from about 119,800 yuan to 990,000 yuan, will begin first deliveries in mid-September.([en.prnasia.com](https://en.prnasia.com/releases/apac/ubtech-launches-uworld-u1-the-world-s-first-full-size-mass-produced-ultra-bionic-humanoid-robot-539322.shtml))
UBTECH books 13k+ preorders for home humanoids
Chinese robotics company UBTECH has launched the UWORLD U1 series, a line of full‑size ultra‑bionic humanoid robots aimed at mass production and home companionship. By July 1, 2026 the company reported more than 13,300 preorders, with prices starting around 119,800 yuan and deliveries planned to begin in September.
Japan pours ¥387B into robot foundation models
On June 30, 2026, Japan’s Ministry of Economy, Trade and Industry said it had selected a SoftBank-led startup, Noetra, and an AIST consortium for a five-year NEDO project to develop a domestic multimodal foundation model for AI robots, budgeting ¥387.3 billion for fiscal 2026. The ministry simultaneously released a revised AI Robotics Strategy targeting deployment of 10 million AI robots in Japan by 2040.
Japan commits ¥10.5T to physical AI robots
On June 27, 2026, MONOist reported that Japan’s new growth strategy foresees ¥10.5 trillion in combined public‑private investment into “physical AI” — robotics and embodied AI systems — by 2040 as part of a broader ¥370 trillion package across 17 strategic sectors. Cabinet Office documents outline investment roadmaps for AI‑driven robots, AI‑grade semiconductors and related infrastructure.
Agility Robotics’ $2.5B humanoid SPAC
Humanoid robotics startup Agility Robotics said on June 24, 2026 it will go public through a merger with Churchill Capital Corp XI valuing the company at about $2.5 billion. The SPAC deal is expected to raise over $600 million to scale production of Agility’s Digit humanoid robots for warehouse and industrial deployments.
Guanglun raises ¥10B for embodied AI data
On June 24, 2026 (07:00 CST), China’s National Business Daily reported that game publisher Giant Network has joined a new ¥10 billion strategic financing round for embodied-data startup Guanglun Intelligent. Earlier coverage from Chinese financial outlets indicates the round includes multiple government-backed funds and industrial investors to expand Guanglun’s physical AI data and evaluation infrastructure.
MIT gives robots language-searchable world memory
On June 17, 2026, MIT researchers unveiled a framework called Describe Anything, Anywhere, Anytime, at Any Moment (DAAAM) that lets robots build long‑term, language‑addressable memory of large environments. By fusing 3D mapping with multimodal vision models, DAAAM allows a robot to attach rich natural‑language descriptions to objects and locations and later answer questions like "Where did I leave my wallet?" based on past observations.
Foxconn Scales Vera Rubin AI Factory Hardware
Hon Hai Technology Group (Foxconn) announced on June 17, 2026 that it is showcasing end‑to‑end AI factory and sovereign AI infrastructure capabilities at the VivaTech conference in Paris. The company highlighted rack‑scale NVIDIA Vera Rubin NVL72 AI supercomputers, Visionbay.ai cloud services and humanoid factory robots as part of its push to lead global AI manufacturing infrastructure.