PaXini AI unveils ONE FOR ALL physical AI loop at Shenzhen open day
PaXini AI on Sept. 7 showcased its ONE FOR ALL physical AI ecosystem in Shenzhen, explaining a PIE strategy behind new products and a 1 billion yuan financing.
Over the past month, PaXini launched the PX6AX GEN4 product matrix equipped with its self-developed GEN4 FUSE true 6D tactile sensing chip; established a Beijing headquarters to form a dual-city model with Shenzhen for research and development as well as manufacturing; and completed a joint-stock reform together with a new financing round of 1 billion yuan. The moves, according to the report, are aligned to the same goal.
At the open day, founder and CEO Dr. Xu Jincheng explained a 'PIE' framework. He said embodied intelligence must solve perception, intelligence and execution in parallel. 'P' (Perception) gives robots physical feedback such as contact, force, sliding and deformation; 'I' (Intelligence) turns each physical interaction into sustainable, reusable and transferable skills; 'E' (Execution) enables robots to complete real tasks using dexterous hands and robot platforms. The three form a causal chain rather than separate product lines: perception determines whether data is real, data determines whether intelligence can evolve, and execution determines whether the technology creates value. This is why PaXini has built a full stack from tactile chips to data, models, dexterous hands and humanoid robots.
Inside the nearly 3,000-square-meter exhibition hall, the six-meter TORA-DOUBLE ONE was displayed. The exhibits followed a path from tactile chips and full-body force perception to data acquisition, dexterous manipulation and multi-robot cooperation. PaXini said the GEN4 FUSE is the world's first and currently the only mass-produced native true 6D tactile sensing array chip, and demonstrated a full-body force sensing suite covering fingertips, palms, skin, joints and feet. The new generation data-collection glove PXCap Pro integrates multi-dimensional tactile sensors, a self-developed miniature high-precision angle encoder and an ultra-wide-angle wrist camera to produce zero-drift pose data and high-fidelity physical ground truth, supporting intelligence evolution.
TORA-series robots demonstrated cluster transport and industrial sorting in a smart-factory setting, as well as reception interactions in a smart-restaurant setting. The report said the demo validated the full causal chain: the GEN4 FUSE converts physical contact into real signals, the PXCap Pro converts human operations into reusable data, models convert data into task capability, and dexterous hands and TORA return that capability to the real world. Every robot movement can be traced to a sensing source, accumulated as data assets and delivered as task results.
During a group interview, Dr. Xu detailed three calculations. The first is the cost of perception: making tactile sensors sensitive is a technology problem, but installing them on every finger, joint and body is an industry problem. PaXini brings semiconductor wafer processes to tactile chip production, which can reduce costs by an order of magnitude and make touch a basic feature of embodied intelligence. The second is data efficiency: what the industry lacks is high-quality physical interaction data aligned with vision, touch, joint pose and operation results. Its OmniSharing DB uses multimodal self-calibration to automate cleaning, calibration and labeling, cutting manual work and token consumption while allowing continuous training and feedback. The third is production-line value: partner BYD will collect complete operation processes on factory lines, turning process experience such as contact, force, operation rhythm and anomaly correction into learnable data. The collaboration aims to strengthen robots' multi-dimensional force-tactile capabilities; BYD sets real problems and production standards while PaXini provides a one-stop answer.
The company, Leiphone noted, is competing for the evolution efficiency of physical AI rather than merely expanding a full-stack portfolio. The 'ONE FOR ALL' concept rests on a closed loop that supports various robots entering the real world—perception decides realness, data drives evolution, and execution creates value.