en.Wedoany.com Reported - Genesis AI announced the launch of GENE-26.5, the first AI foundation model that equips robots with human-level physical manipulation capabilities. The company simultaneously released a first-of-its-kind system designed to unlock massive amounts of data and train the GENE-26.5 robot foundation model at scale. The system includes two proprietary components: a life-sized dexterous robotic hand that enables direct skill transfer from humans to robots, and a high-performance data engine, thereby resolving the data bottleneck that has long constrained robot foundation models.
The GENE-26.5 robot foundation model is purpose-built for the robotics domain, capable of absorbing vast amounts of data and environmental information, enabling robots to perform long-duration, highly complex tasks. Genesis AI released a video showcasing a series of operations previously difficult for robots to achieve, such as cooking a 20-step meal, making a milkshake, conducting high-precision laboratory experiments, wire harness bundling, solving a Rubik's Cube, single-hand multi-object grasping, and human-level piano playing, with dexterous movements and smooth, coordinated bimanual operation.
For a long time, the morphological differences between humans and robots have limited the effectiveness of learning from human data. Genesis AI's dexterous robotic hand, combined with a data collection glove equipped with tactile electronic skin, achieves precise mapping among the glove, human hand, and robotic hand, allowing everyday human operations to be directly converted into high-quality training data. The hardware cost of this glove is a hundred times lower than conventional solutions, and internal tests show its data collection efficiency can be increased by up to five times with higher quality, making continuous, large-scale robot training feasible. The company is working with partners to deploy the gloves in actual workplaces, turning daily work into new data sources, while also incorporating data such as first-person video and internet video to help build a large-scale human skill library.
At the simulation level, Genesis AI significantly narrows the gap between virtual and real environments through high-fidelity rendering and physics simulation, allowing robot performance in simulated tasks to closely approximate real-world operation. This simulation system supports the self-evolution and iteration of the entire technology stack in a fully virtual environment, greatly improving training and evaluation efficiency and avoiding the limitations of traditional physical testing, which involves long cycles, high costs, and difficulty in scaling.
The company stated that its first general-purpose robot is forthcoming, having already integrated the technology announced today. Genesis AI has completed a $105 million seed funding round, with investors including Eclipse, Khosla Ventures, Bpifrance, and HSG, among others. More details can be found at genesis.ai.
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
