PHANES AI and Tinno Mobile Launch SeePhys Hub for Full-Body Multimodal Data Collection

The hardware consists of a four-lens headband, two wrist bands, three waist and lower-leg motion trackers and one tactile glove. Together they capture a 360-degree first-person view, hand pose tracking with high-density tactile sensing, and body pose and force data.

The release follows a shift in robot data collection away from body-bound teleoperation toward lighter devices. Leiphone describes the path as running from teleoperation to UMI and then to Ego, with equipment becoming lighter and recording simpler. PHANES argues that the simplification has dropped information that matters: Ego-style capture can log what the hands are doing but has difficulty recording whole-body motion and cannot directly measure the pressure applied when a hand touches an object. Those signals, the company says, are often what determines whether a task is completed.

Estimates cited in the report put the requirement for Ego data at around 100 million hours. PHANES takes the position that collection hours do not translate directly into learning gains, since data quality, the match with robot tasks and the training method also affect results. The company says it wants to test whether synchronously recording body motion and touch alongside vision produces more learning value per hour, and whether full-modality data in the millions of hours can support the expected generalization. As an example, the report cites bed-making: a person grips a corner of the sheet, lifts an arm to spread it, then moves along the bed to smooth the creases. Body position determines what the hands can reach, while grip location and force shape how the fabric unfolds, and a vision-first Ego recording captures neither the torso and leg motion outside the frame nor the contact pressure during pulling.

SeePhys Hub is the first engineered product from a research line that previously included EgoTouch, TouchAnything and TouchWorld. EgoTouch synchronized first-person video, hand pose and pressure distribution; TouchAnything learned to estimate contact location and pressure from vision in order to expand tactile data; TouchWorld brought touch into robot action, predicting a contact target before a movement and correcting the action using real feedback during execution. The new system turns that work into a wearable rig: the headband's four cameras record the scene, wrist views capture close-range manipulation, trackers on the waist and both shins record body movement, and the tactile glove collects contact pressure from both hands.

The headband weighs no more than 220 grams and each wrist band no more than 75 grams, and both support 24 hours of continuous collection, according to the report. Synchronization error among the headband's cameras, IMU and UWB is kept within 0.5 milliseconds so that vision and motion data from the same moment can be aligned. A self-developed processing platform handles parsing and cleaning, splits continuous operations into action segments with semantic annotations, and connects scene reconstruction, full-body pose estimation and hand motion tracking. Motion retargeting maps human movement onto robot bodies for later policy training and hardware adaptation.

PHANES AI was founded by Yang Shuo, a tenured professor and doctoral supervisor at the School of Computer Science and Technology at Harbin Institute of Technology, Shenzhen. The company's stated longer-term goal is a world model for whole-body mobile dexterous manipulation, in which body motion, hand contact and environmental change are modeled together so a robot can predict the outcome of moving, adjusting posture or changing contact and revise its judgment using execution feedback. The company says more than 90 percent of its core team members graduated from Tsinghua University, Peking University, the University of Hong Kong and Harbin Institute of Technology, with backgrounds spanning multimodal large models, world models, large-scale data processing and hardware.

【Editor's Summary】PHANES AI and Tinno Mobile have released SeePhys Hub, a wearable rig that records first-person vision, hand touch and full-body motion in one session and feeds the data into a processing pipeline for robot training. The product marks the company's move from tactile-sensing research into an engineered data-collection device, aimed at building world models for whole-body mobile dexterous manipulation.