en.Wedoany.com Reported - The latest evaluation results from RoboChallenge, the world's first large-scale real-machine evaluation platform for embodied intelligence, were announced on May 20. The Era0 model from Chinese company RobotEra achieved a composite first-place ranking in the Table30 series tasks, with a success rate of 64.33% and a process score of 76.34, and topped all 17 desktop dexterous manipulation tasks. What drew even more industry attention was that in the two highest-difficulty tasks—"making a vegetarian sandwich" and "wiping the table"—where all previously evaluated models had ended with a zero success rate, Era0 became the only model on the leaderboard to achieve a non-zero success rate.
The RoboChallenge platform, where Era0 was evaluated, is known in the industry as the "real-machine examination hall" for embodied models. The platform was jointly launched by Dexmal and Hugging Face in October 2025. Its test cluster covers 20 mainstream robots, including UR5, Franka, ARX5, and ALOHA, conducting "blind tests" on all participating models under unified, standardized hardware constraints. Its Table30 task set covers 30 standardized desktop dexterous manipulation tasks, with the entire testing process and results being open and traceable end-to-end.
RobotEra's Era0 ranked first across all dimensions of Table30, including bimanual coordination, flexible object manipulation, multi-view perception, classification tasks, and long-horizon tasks. Besides Era0, other global leading embodied models competing on the same leaderboard included Physical Intelligence's π0/π0.5, Microsoft's CogACT, as well as models from Chinese developers such as Dexmal's DM0 and Spirit AI's Spirit-v1.5. It is worth noting that RoboChallenge has been officially incorporated as a competition into two major international top-tier conferences, CVPR 2026 and ICRA 2026, becoming an important benchmark for measuring the real-world deployment capabilities of embodied models.
The 64.33% success rate achieved by RobotEra's Era0 in this evaluation significantly surpassed the second-place Dexmal DM0 at 62.00% and the third-place GigaBrain-0.1 at 51.67%, and far exceeded π0.5's 42.67% and π0's 28.33%. For industry reference, the RoboChallenge annual report released in February 2026 showed that the average success rate of the top model on Table30 at that time was just slightly over 40%, with success rates for fine manipulation tasks generally below 10%. Four months later, Era0 has dramatically raised this benchmark to 64.33%, with breakthroughs being most evident in flexible object manipulation and long-horizon memory tasks.
A relevant person in charge at RobotEra stated that Era0 achieves deep integration of the Vision-Language-Action (VLA) model with a world model, forming synergistic breakthroughs across data, model, and inference. On the data side, the team established quantified quality inspection Standard Operating Procedures (SOP) and a cross-embodiment pre-training dataset, ensuring data quality and model transferability. On the model side, the introduction of visual grounding perception enhancement technology enables the robot to complete highly generalizable challenges such as "throwing a white paper ball into a pure white trash can in an all-white environment"; simultaneously, a short-range temporal memory mechanism was introduced, allowing the model to remember recent actions and states when executing long-horizon tasks, avoiding repetitive errors and circling caused by relying solely on current frame information. On the inference side, techniques such as action interpolation, parameter adaptation, and gripper post-processing improved trajectory continuity and execution stability.
RobotEra was founded in August 2023, incubated by the Institute for Interdisciplinary Information Sciences at Tsinghua University. It is the only embodied intelligence enterprise in which Tsinghua University holds equity. Founder Chen Jianyu is a doctoral supervisor and assistant professor at Tsinghua University. The company adheres to full-stack self-research integrating both software and hardware for the "embodied brain" and "humanoid body," with over 95% of core hardware developed in-house, building a product matrix of "humanoid robots + core components + scenario-based solutions." In July 2025, the company completed a Series A funding round of nearly 500 million yuan, followed by a Series A+ round of nearly 1 billion yuan in November of the same year. Cumulative order value has exceeded 500 million yuan, with deep collaborations established with companies such as Geely, Renault, SF Express, TCL, Haier, and Lenovo.
Era0's technological advantages have extended from evaluation leaderboards to the front lines of industry. Currently, RobotEra has engaged in deep cooperation with leading logistics enterprises such as China Post and SF Group. Related technologies and products have achieved regular operations in over 10 logistics centers across 5 provinces and cities nationwide, achieving the industry's first Product-Market Fit (PMF) implementation for embodied intelligence products in the logistics sorting sector. In warehousing logistics scenarios, the efficiency of robotic logistics parcel feeding and pharmaceutical sorting operations has reached 70% of the human level, with the largest single order in the logistics industry approaching 50 million yuan. In the automotive manufacturing sector, the company focuses on scenario tasks such as component picking, high-precision assembly, and quality inspection, and is advancing scenario validation with multiple OEMs.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
