Third round in a month
Shutu Technology recently closed a Pre-A round of nearly RMB 100 million, co-led by CDH Investments’ BaiFu arm and the Kunpeng Fund, a national-level data fund, with Guangzhou Industrial Investment and Yihe Capital participating and an early backer, Trend Investment, topping up. It is the company’s third financing within a month. On top of its delivered video-upscaling data product, the round funds research and validation of tactile-data representation training and accelerates the SynaTac tactile product line and edge-deployment business.
From landing backward to training infrastructure
Building on its delivered and validated data business, Shutu extends through SynaTac toward contact intelligence, forming a dual-engine matrix with SynaData for world understanding and contact execution. It also runs embodied-data and skill benchmarking, linking training-data supply, contact-strategy research and real-task validation. The chief technology officer, Dr Wei Mu, was previously vice-president and China software lead at Covariant AI, and Shutu’s core team is built from Covariant’s algorithm group, with deep experience from demo to product to field. Having practised the path from embodied demo to stable field operation, the team knows the gap between can demo and can deliver is not just algorithmic but a continuous iteration of data, training and validation around field problems.
From the field problem, Shutu argues the core gap in embodied-algorithm landing is insufficient world understanding at the brain layer and unstable contact execution at the cerebellum. Massive human-action video-upscaling data helps the robot understand the task, the know-how. Tactile-representation modelling’s value is reliable execution, the do-it-well. An embodied model must judge whether contact is stable and adjust force and motion against slip, deformation and resistance in real time. More sensors and more signals do not equal contact ability. The key is making that information actually enter learning and operation.
The dual engines
Shutu has built two human-behaviour learning engines for world understanding and contact execution. SynaData, the mature product base, turns human operation video into structured multimodal training data covering action trajectories, object poses and contact, serving more than 60 per cent of RMB 10 billion-valued embodied-AI unicorns with bulk purchases from top tech vendors. SynaTac builds on unified structured tactile representation, aligning object, action and contact state and fusing vision, touch and robot state, then trains contact strategy through demonstration, interaction and trial-and-error with high-frequency feedback for edge deployment on dexterous hands and different bodies.
The first public embodied-data and skill benchmarking system links data supply, model training and robot execution across three stages: data-metric evaluation, trainability evaluation, and task and real-environment validation, feeding results back into production, representation, training and skill iteration.
Editor’s note: This is an adapted translation of the original Leiphone report. It has been trimmed and restructured for readability for an international business audience.