Ropedia Raises Millions for Physical AI Real-World Data

Ropedia, a Singapore-based startup, has successfully closed an angel funding round, securing tens of millions of dollars (equivalent to several billion Japanese Yen). The investment saw participation from a diverse group of venture capital firms, alongside strategic contributions from companies specializing in smart manufacturing and AI model development. This capital infusion is earmarked for bolstering its core technical team, scaling hardware production, and fulfilling large-scale data product deliveries.

The company is addressing a critical bottleneck in the rapidly advancing field of physical AI: the scarcity of high-quality, real-world data. Unlike traditional AI models that learn from static datasets, physical AI for robotics requires a rich, multi-dimensional understanding of the environment. This includes not just visual input, but also an awareness of physical dimensions, dynamic interactions, object manipulation, movement trajectories, and contextual scene understanding. Current data acquisition methods are often prohibitively expensive, requiring specialized equipment and complex setups, followed by extensive processing to transform raw data into usable formats for AI training.

Ropedia's innovative approach centers on its proprietary 'Experience Engine', a data engine designed to create a cost-effective foundation for real-world data. This engine facilitates an end-to-end data production pipeline, encompassing collection, refinement, training, and validation. It automatically converts various forms of raw data into what the company terms 'real-world experience points', directly applicable for AI model training and evaluation. Their recently launched 'HOMIE' data collection system, a head-mounted device, captures first-person perspectives of human actions, environmental changes, and object interactions.

The significance of this data is underscored by the company's 'Xperience-10M' dataset, an open-source release featuring approximately 10 million instances of human behavior and environmental interactions, approaching one petabyte in size. This dataset has already been instrumental in the research and training of prominent AI models, including Alibaba's 'Qwen-VLA' and Allen Institute for AI's 'MolmoMotion', accumulating over 2.7 million downloads. Ropedia currently serves over 1,800 data utilization organizations, with more than 500 active data usage agreements, and counts over 20 embodied AI manufacturers and IT firms among its clientele.

Evidence of the data's impact is emerging. In one instance, the 'ACE-Ego-0' model from ACE Robotics utilized 435.7 hours of first-person video from Xperience-10M. Ablation studies demonstrated that models trained with a combination of robot data and human video achieved a task success rate of 72.8%, a 4.5 percentage point improvement over models trained solely on robot data. The company has experienced substantial growth, with revenues increasing approximately fivefold in the past six months, driven significantly by the North American market which accounts for 60-70% of its revenue.

Looking ahead, Ropedia is developing a next-generation multimodal data collection system set to launch in August 2026. This system aims to more comprehensively integrate first-person visual data, hand movements, human motion, and spatial positioning. CEO Chen Zhaoxi emphasizes that the future of multimodal data lies not in simply increasing sensor count, but in synchronizing visual, motion, tactile, and spatial data within a unified temporal and coordinate framework to accurately replicate physical movements for AI learning. This is crucial for tasks requiring nuanced physical understanding, such as assembly or intricate manipulation, where tactile feedback and force application are critical differentiators often missed by visual-only data.