A single destination for timely, editor-curated robotics news from around the world.
In October 2026, SuperCLUE conducted an office-agent benchmark test, awarding Doubao Work an overall score of 85.33. Kimi Work and Tencent WorkBuddy followed closely, showcasing competitive performance in document handling, writing, and data tasks. This ranking highlights the advancements in office...
AGIBOT has introduced GO-3 Preview, an embodied foundation model that enhances robots' learning capabilities throughout their lifecycle, from pre-training to post-deployment. This model has achieved first-place rankings on three significant embodied AI benchmarks: RoboDojo, RoboTwin, and RoboColiseu...
RoboColosseum has launched an open initiative aimed at testing robot AI on physical hardware, providing access to over 100 robots and 140 real-world scenes for academic researchers. Supported by Unitree, the initiative features its G1 platform in the humanoid category and includes participation from...
X Square Robot has launched X2Real, a new simulation benchmark designed to align simulated robots with actual hardware performance. This innovative tool encompasses 44 tasks across 10 categories and employs AI agents to generate new testing scenarios. The company claims a high success correlation of...
Robotera's Video Prediction Policy 2 (VPP2) has secured the top position on RoboDojo's simulation leaderboard, announced on October 9. This model effectively links scene change predictions with corresponding robot actions, outperforming PhysicalRSI in both average score and task success rate. VPP2 a...
Humanoid robots were tested in challenging weather conditions during the National Day celebrations, demonstrating their capabilities in real-world scenarios. This testing marks a significant advancement in embodied intelligence, showcasing how these robots can operate effectively in adverse environ...
OpenAI's GPT-6 Astra achieved the highest overall score in SuperCLUE's September 2026 EmbodiedCLUE-VLA evaluation, leading with a score of 85.71 in interaction and planning. This score significantly surpasses the runner-up, Gemini-3.8-Flash, which scored 64.29, highlighting Astra's notable advantage...
Vals, an AI benchmarking startup founded in 2024, has successfully raised $40 million in a Series A funding round led by Andreessen Horowitz. This follows a previous seed round backed by 8VC and Bloomberg Beta. Co-founder Rayan Krishnan highlighted that Vals aims to address the shortcomings of tradi...
An open-source benchmark has been developed to assess the capabilities of AI agents in engineering functional robots. This initiative comes in response to the increasing automation of software coding, where AI systems can autonomously write and revise programs. The significance of this benchmark li...
DeepCybo has announced the open-source release of PhysBrain 1.5, achieving an impressive average score of 72.5 across 28 public embodied benchmarks. This model, which features both 2B and 8B parameters, is designed to enhance unified understanding, action, and future-state token capabilities, alongs...
Robocurve has successfully raised $10 million in seed funding aimed at expanding its independent testing of frontier AI models in controlling real robots. The funding round was led by Initialized Capital, with participation from several investors including Notable Capital and Y Combinator. Robocurve...
OpenAI's GPT-6 Astra has achieved a remarkable 95% success rate in a gross pick-and-place task using dual I2RT YAM robotic arms, according to evaluation data from RoboCurve. This performance highlights Astra's advancements in spatial reasoning and physical manipulation, surpassing competitors like A...
Baidu has unveiled DuMateBench, a new evaluation leaderboard aimed at assessing the ability of AI agents to perform real-world tasks and produce usable outputs. This benchmark encompasses over 200 office tasks categorized into six distinct areas, challenging agents in complex operational settings....
Shenzhen's X Square Robot achieved a remarkable feat by sorting 1,816 parcels in a single hour using basic grippers. This performance surpassed its self-imposed target of 1,248 parcels, which aligns with Figure AI's sustained hourly average. This accomplishment is significant as it highlights the c...
AGIBOT announced that its WITA-Omni Preview multimodal foundation model has achieved the highest score on the Daily-Omni audio-visual reasoning benchmark, surpassing models from Alibaba, Google, and ByteDance. The model recorded an average accuracy of 85.21 percent, excelling in audio-visual alignme...
Ant Bailing has announced the release of its Ling-3.0-flash model, which features a total of 124 billion parameters and 5.1 billion activated parameters. This model has achieved impressive results, securing 15 first-place and 19 second-place rankings across 34 evaluation dimensions, tying with DeepS...
The development of robotics begins long before physical assembly, relying heavily on simulations to validate designs and refine algorithms. These simulations demand significant computational resources, making system benchmarking crucial to identify hardware limitations early in the process. By measu...
Recently, Zhiyuan has established a new standard for the capabilities of work robots, showcasing their potential in various applications. This development is significant as it highlights the evolving role of robots in both industrial and domestic environments, pushing the boundaries of automation....
AGIBOT has introduced the GO-3 Preview, an advanced embodied foundation model that enhances learning throughout the entire lifecycle of robot training and deployment. This model has achieved first-place rankings on three significant embodied AI benchmarks: RoboDojo, RoboTwin, and RoboColiseum, showc...
WorldOracle has introduced a benchmark to assess action-conditioned world models (ACWMs) by comparing their predictions with actual robot executions. In a demonstration, a robot's grasp failed while the model inaccurately predicted success, highlighting discrepancies between simulated and real-world...
UBTECH has announced that its Thinker-WM has achieved top scores in three embodied world model benchmarks. The integration of depth data significantly improved single-task success rates on the RoboTwin platform, increasing performance by 15 percentage points during ablation tests. This advancement...
Elorian and Scale AI have launched a new benchmark called Humanity's Sixth Sense (HSS) to evaluate AI models' intuitive visual reasoning capabilities. The benchmark consists of 522 questions based on 288 images and 234 short video clips, covering various reasoning categories. While human participant...
The HiPHI dataset, comprising 617.5 hours of whole-body human motion data, aims to enhance the training of humanoid robots. Captured with optical motion capture technology at sub-millimeter accuracy, it includes 245.7 hours dedicated to human-object interactions, providing synchronized object trajec...
In-home assistive robots are being developed to clean and organize homes while owners are away. However, without improved memory capabilities, these robots may create more issues than they resolve. For instance, if a robot misplaces important items like wallets and keys, it could lead to significant...
Simate, a newly established company, has introduced Simate-beta, a rapid-response physical AI model. This model has achieved a remarkable average score of 33.95 on the RoboDojo benchmark, indicating its superior performance in the field. The significance of this development lies in Simate's innovat...
ZDTaichu 5.0-9B, a 9 billion parameter multimodal model, has achieved remarkable results in spatial embodied intelligence, securing first place in 8 out of 9 international benchmarks. It outperforms competitors like Qwen 3.5-9B and STEP 3-VL-10B in spatial perception and three-dimensional reasoning...
Dr. Scott Walter has proposed the Humanoid Decathlon Challenge, aiming to establish a comprehensive benchmark for humanoid robots. This initiative comes in response to recent athletic achievements in robotics, where humanoid robots have excelled in isolated tasks but lack a unified performance stand...
On August 20, OpenRouter introduced stealth/ox-alpha, an anonymous AI model that has demonstrated superior coding capabilities compared to several closed frontier models. This unexpected launch has ignited speculation within the industry regarding the identity of the model and its implications for A...
The University of Hong Kong's Multimedia Laboratory has developed 'RoboDojo,' a comprehensive platform for assessing robotic manipulation in both simulated and real-world settings. This initiative, led by Professor Ping Luo and Ph.D. student Tianxing Chen, involved collaboration with researchers fro...
Unitree's recent IPO, priced at 150.80 yuan with a market cap of 60.99 billion yuan, has established a new valuation standard for Chinese humanoid robot companies. This significant financial milestone will influence how other firms, including DeepRobotics, Leju, DOBOT, AgiBot, Zhongqing, and X Squar...
Spirit AI's Spirit v1.6 temporarily surpassed Nvidia on the RoboArena robotics benchmark, highlighting the fierce competition between the US and China in AI development. However, this achievement was short-lived as the benchmark's creators revised their methodology and subsequently removed Spirit's...
At the WAIC 2026 event, Muka Robotics, Shengshu Technology, EvoPhys.ai, and Chengwei Capital highlighted a significant shift in the global tech landscape. They noted that Chinese world model startups are now operating independently, without American counterparts to benchmark against, which challenge...
AgiBot WITA-Omni has achieved a score of 85.21 on the DailyOmni benchmark, securing first place in 6 out of 8 indicators. This performance surpasses notable competitors such as Google Gemini, ByteDance Doubao, and Alibaba Qwen in the realm of embodied cross-modal understanding. The significance of...
The BYD AI team has introduced the HyWorldVLA, a hybrid pixel-latent world model that utilizes VLA architecture. This model has achieved a score of 90.59 PDMS on the NAVSIM benchmark, marking a significant milestone for BYD in the field of autonomous driving foundation models. This achievement is n...
Epoch and METR have introduced MirrorCode, a benchmark designed to assess AI systems' capabilities in long-horizon programming tasks. The benchmark, which was announced in April, has shown that AI models, such as Opus 4.7, can complete tasks that would typically take humans weeks in a fraction of th...
A recent evaluation by the UK Artificial Intelligence Security Institute and the US Center for AI Standards and Innovation revealed that Moonshot AI's Kimi K3 scored 32.2% in offensive cybersecurity capabilities, significantly lower than the average score of 76.2% achieved by leading US models. This...
Starmind's orbital compute technology presents a significant advantage over traditional ground-based data centers by eliminating constraints related to land, water, and grid permitting. While terrestrial data centers are currently cheaper and faster to construct, with U.S. data center spending reach...
NVIDIA has unveiled its Nemotron 3 Ultra, a new AI orchestration platform that promises superior performance at a more affordable price compared to leading closed models. This innovative system has been optimized by LangChain, which has fine-tuned its Deep Agents to work seamlessly with the Nemotron...
RLWRLD, a physical AI company with a proprietary robotics foundation model, RLDX-1, has announced the launch of “All Hands Up!”, an open web platform that provides technical reports and visualization tools based on the company’s firsthand experience operating a wide range of commercially available d...
Lumos Robotics says its Prime R0 industrial embodied AI model has achieved the highest overall score on the latest MolmoSpaces leaderboard, outperforming larger models from competitors including Nvidia and research teams from the United States. The Chinese robotics company said its 2.8-billion-param...
Chinese medical AI has achieved significant advancements, highlighted by the entry of a teleoperated surgical robot into the European Union market and a clinical-grade model surpassing a key healthcare benchmark established by OpenAI. On Monday, Shanghai MicroPort MedBot announced that its Toumai Re...
Artificial Analysis has launched AgentPerf, the industry's first benchmark for agentic AI, aimed at providing developers, enterprises, and infrastructure providers with a standardized method to evaluate and compare various systems designed for agentic AI capabilities. The initial results have highli...
Waymo has developed an advanced computer model aimed at enhancing its understanding of human behavior during crash scenarios involving its robotaxis. This initiative, which comes as part of the company's ongoing efforts to improve safety and reliability, utilizes extensive data analysis to simulate...
Daimon Robotics and Galbot have announced the launch of RobOmni, a new platform designed to benchmark tactile perception and dexterous manipulation in the field of embodied AI. This development marks a significant shift from traditional vision-centric approaches to a more comprehensive understanding...
Kuawei Intelligence, a leading Chinese company in embodied artificial intelligence, has secured the top position in the WorldArena Track 2 (Data Engine) global benchmark for May 2026. This accomplishment places Kuawei ahead of notable international competitors such as WoW and BLM. The ranking not on...
At the NVIDIA GTC Taipei 2026 event, a significant milestone was announced for China's embodied intelligence sector. This achievement highlights the advancements made in the field, showcasing the country's commitment to innovation and technology development. The announcement is expected to bolster C...
The National Institute of Standards and Technology (NIST) has introduced a standardized performance benchmark and testing procedures aimed at assisting developers and evaluators of humanoid robots. This initiative is designed to establish a consistent framework for assessing the capabilities of huma...
A consortium of Chinese research institutions has unveiled RoboMemArena, marking the introduction of the first comprehensive benchmark designed to assess robotic memory in long-horizon manipulation tasks. This initiative aims to enhance the capabilities of robots in performing complex tasks that req...
A new standardized evaluation system for robot memory, known as the RoboMME benchmark, has been introduced by a collaborative effort involving Michigan University, Stanford University, and Figure AI. This innovative framework assesses robot memory across four key dimensions: temporal, spatial, objec...
Fraunhofer IPA has introduced a new benchmark aimed at evaluating humanoid robots for industrial applications. This initiative addresses the need for standardized criteria that can be utilized by third-party analysts to assess the performance and suitability of these robots in various industrial set...
RSF defines a common language for robot service capability, lifecycle operations, certification pathways, and service-provider networks.
Daily robotics news, in-depth analysis, conference highlights, and discussions with professionals worldwide.
